Industrial Robot With Controller

The present invention relates to industrial robots or computer-controlled manipulators and, more particularly, to the design and control of an articulated mechanical arm of an indeterminant number of axes, capable of being configured with kinematic redundancy.

Industrial robot arm designs have followed a few basic types. Industrial robots can be classified according to their mechanical linkage geometries, i.e., the particular arrangements of structural elements and joints which connect them and the associated motion control systems required to coordinate joint action to produce straight line motion and other controlled paths at the toolpoint. In the most general purpose and versatile manipulators, six degrees of freedom are incorporated in the linkage configuration to provide complete control of the position in space and orientation of the tool mounted at the end of the manipulator.

One elementary form of manipulator employs a set of three slides connected at prismatic or sliding joints. These slides are disposed in a nominally orthogonal arrangement to position the "wrist" of the device, a second set of three orthogonally disposed rotational axes which determine tool orientation. This mechanism geometry provides a roughly rectilinear working volume. Such a device is typified by the IBM RS 1 robot. The "cartesian" geometry of such a device has a number of distinct advantages over other types. Most important of these is that no coordinate transformations are necessary to produce useful controlled motions at the toolpoint. Instead, linear and circular interpolation of the positioning axes is sufficient.

A second common mechanical geometry provides a wrist, as described above, linked to two slides which are disposed at a right angle and connected at a prismatic joint. These slides are affixed to a revolute joint in the base of the device which provides rotation about a vertical axis. The workspace of such a manipulator is roughly "cylindrical" in shape. Such a device is typified by the Prab Model FA robot.

In a third common mechanical geometry, the wrist described above is positioned in space by a slide connected at a prismatic joint to a revolute joint which, in turn, is mounted at a right angle to and rotated around a vertical axis by a second revolute joint in the base of the device. Theoretically, this type of "polar" geometry produces a spherical work space. In practice, mechanical design considerations generally restrict the useful workspace to a spherical shell less certain significant conical sections. Such a device is typified by the Unimation Unimate 1000 robot.

More sophisticated motion control systems are required for arms with cylindrical and polar linkage geometries than for arms with cartesian geometries because coordinate transformations must be performed to generate straight line movements at the toolpoint. However, as a class, manipulators which employ one or more slides connected at prismatic joints exhibit certain significant performance limitations. These are due, in part, to the relatively large size and high weight, and the resulting high motive power required of such a design to convey and locate in space a tool or workpiece of a given mass. They are also due to the fact that the positioning slides often interfere with other objects in and around the working area, including the workpiece itself.

Of the manipulator types described above, the cartesian systems tend to be the least spatially efficient linkage configurations, since the workspace is often completely surrounded by a large framework of positioning slides and supporting structure. Of the group described above, the polar type, which employs only a single slide, is the most efficient and least intrusive in the workspace. To minimize the spatial efficiency problem, a few polar geometry devices have been designed in which the slide collapses upon itself when retracted to minimize interference problems. In one form, for example, a set of colinear slide segments telescope. This form is characterized by the Maker, which is manufactured by U.S. Robot. In another form, a thin-wall steel tube that forms the slide when extended is caused to collapse in section to a flat sheet which can be rolled onto a drum when retracted. This device is typified by the Martin Marietta/NASA Viking Lander arm. Mechanical implementations of these designs tend to have relatively poor static and dynamic performance characteristics, however, due either to the number of additional prismatic joints incorporated to provide telescoping or to the very thin-wall slide cross-section.

To improve the performance and workspace interference characteristics of manipulators, a linkage geometry which permits considerably more efficient mechanical designs has been devised in which a series of rigid link segments connected by revolute joints is used to position the wrist. This is known as a revolute or jointed arm manipulator and is the type of the present invention. In a general-purpose manipulator of this type the wrist positioning mechanism typically consists of two links connected by a revolute joint, with a terminal end of one of these links mounted on a second revolute joint fixed in plane with the first, itself mounted at a right angle on and rotated about a vertical axis by a third revolute joint in the base. Manipulators which employ this linkage geometry are more like the human arm than the earlier designs described above, but function kinematically more like a "backhoe" than a human limb, since the linkage configuration operates in a fixed plane. Theoretically, such a jointed arm linkage geometry produces a spherical working envelope. Like cylindrical and polar geometry manipulators, the jointed arm manipulator requires a relatively complex controller which must perform coordinate transformations to produce straight lines or other controlled path behavior at the toolpoint. The principle advantages of the jointed arm manipulator geometry relate to the fact that when the arm linkage which positions the wrist is retracted, it folds upon itself, permitting arms to be relatively compact for a given working envelope and light-weight for a given payload.

Two distinct mechanical embodiments of the jointed arm geometry have gained acceptance in the industry. In one, the actuators which drive many of the arm and wrist joints are mounted some distance from the joints themselves. In such designs, motors and gear reducers mounted at the "shoulder" transmit power to the joints through the effect of a four-bar linkage configuration or through pushrods and bellcranks, or by chains, timing belts, or other "tendon" arrangements. An example of such a device is the ASEA IRb 6 robot. This design has the advantage that the relatively bulky and heavy motors, drives, and velocity feedback hardware need not be packaged with and supported by the more distal arm structure. Consequently, motive power requirements for a given payload may be reduced. Nevertheless, the drive train which is employed to transmit power to a remote joint itself imposes a number of significant performance limitations. The feasible range of motion of joints is often limited by geometric ratio changes or over-center conditions in the power transmission mechanism, resulting in arm assemblies with relatively restricted, toroidal working envelopes. Transmission mechanisms also add considerable inertia, compliance and mechanical inaccuracy to the drive train, to the detriment of static and dynamic performance. Moreover, since the transducer which is used to determine toolpoint position is often mounted at the origin of the transmission, compliance and mechanical inaccuracy in the transmission significantly reduces the precision of the device.

In the second common embodiment of the jointed arm linkage geometry, substantially all of the actuators which drive the arm and wrist joints are located on or within the arm structure adjacent the joints. In some cases, actuators are located directly at the joints; in other cases, they are located in adjacent "in-board" link segments. This arrangement overcomes the problem of limited joint travel and, as a result, certain mechanisms of this type exhibit useful working envelopes that approach a sphere. The joint-mounted or link-mounted drive design also reduces or eliminates problems associated with power transmission inertia and compliance. An example of such a device is the Unimation PUMA 600 robot.

Although jointed arm geometry provides a more efficient operation than cartesian or polar configurations in terms of maneuverability, working envelope, and overall dexterity, it requires a more sophisticated controller, capable of performing elaborate and time-consuming coordinate transformations to position the toolpoint. The relative complexity and high cost of computer control systems required for accurate and, to an increasing extent, adaptive control of the toolpath in jointed-arm manipulators have had a significant influence on the particular linkage geometries utilized in most commercial arm designs. Linkage designs have generally been adopted which simplify the process of coordinate transformation and reduce the number and rate of computations that must be performed. For example, common jointed-arm linkage geometries avoid "off-set" pitch joints, a feature which greatly complicates transformation. Thus, by imposing specific constraints on the linkage geometry, explicit mathematical expressions (i.e. closed form, analytic expressions) can be obtained for the coordinate transformations which simplify the control system.

However, such efforts to constrain mechanical design for the sake of control system efficiency exhibit several shortcomings. The linkage geometries that allow for explicit solutions to the transformation equations often are not optimal for performance and cost. In addition, explicit expressions can not be easily adjusted for mechanical imprecisions in the manipulator. Moreover, it is doubtful that any closed form solution exists for the transformation equations of any redundant manipulator. Furthermore, intrinsic to all jointed arm manipulators is a condition known as a singularity. Conventional control systems and manipulators will encounter regions in their working envelope containing "singularities" which prevent effective operation of the controller. A singularity is the condition of the loss of one or more degrees of freedom of movement in one direction and redundancy of movement in another. Singularities are encountered in the work envelope when the manipulator assumes particular positions. When a singularity is encountered by a manipulator, one or more of the joints of the machine is effectively eliminated from contributing to the movement the manipulator is directed to execute. Thus, when a singularity is approached, a six-axis manipulator is reduced to five degrees of freedom of movement, a fiveaxis machine to four, etc. Conventional controllers are unable to operate efficiently when sets of singularities are encountered in the work envelope because the equations typically used to control motion have no mathematical solution at a singularity. Thus, while many mechanical designs have been influenced by an effort to simplify the mathematics associated with the control of the manipulator, mathematical control problems persist.

Moreover, as the mechanical design and linkage geometry improve the versatility of the manipulator, the problems of singularities are increased. Where these problems have been confronted in the current technology, the approach typically has been to anticipate and trap the singularity conditions. That is, the controller is programmed to anticipate the impending occurrence of a singularity and special routines are provided to avoid near singular conditions. However, such a technique fails to take advantage of the fact that a redundant manipulator arm, by its very nature, has an advantage over singularities if properly controlled. Conventional manipulators fail to take advantage of the benefits provided by the use of seven or more axes of movement because the controllers of such robots are unable to deal effectively with the complexity introduced in the control mathematics by the presence of seven or more degrees of freedom. In accordance with the present invention, it has been found to be better to adopt an optimum mechanical design, unconstrained by concerns about mathematical complexity, and to confront the mathematical problems of coordinate transformation, redundancy, singularities and mechanical imprecisions by the adoption of iterative control methods.

In addition to the four basic types of general-purpose manipulator described above, each of which provides six degrees of freedom at the toolpoint, many other linkage geometries have been devised for special applications. In the design of most of such special-purpose arms, an effort is made to employ a linkage geometry having the minimum number of driven joints necessary to perform the particular task of that application. Significant cost savings result from such an effort through a reduction both in the number and size of structural components, motors, power supplies, servo feedback hardware and in the complexity of the control system required. Special-purpose manipulator designs have evolved, for example, for the relatively simple kinematic function of loading and unloading workpieces from lathes. One type employs a two-axis, cartesian mechanism in which the primary slide is mounted parallel to the lathe spindle centerline. A second common type uses two links connected at revolute joints, plus one short slide, to handle short chucked parts. Because of their uniquely tailored mechanical designs, neither of these specialized manipulators requires the controller to perform coordinate transformations. In both, the orientation of the linkage geometry itself produces appropriate tool paths for the given application when each joint is driven independently in the proper sequence. With such designs it is possible that the joints may not require an analog servocontrol network.

As a second example, it has been determined that for a large class of MIG welding operations in the factory, control of rotation of the welding tip about its axis is not necessary and consequently, that a three-axis arm with a two-axis wrist provides sufficient tool control. Many other examples exist of manipulator designs being optimized for a specific task or class of tasks. In most cases, a geometry providing less than six degrees of freedom at the tool is employed and the physical sizes of the links and/or slides, as well as their load capacities, are matched to the specific application. Accordingly, a unique design is required for each such specialized application.

Previous manipulator designs exhibit a number of significant limitations and shortcomings in addition to those mentioned above. Jointed arm manipulators, incorporating six revolute joints and providing six degrees of freedom at the toolpoint, while more efficient than other general-purpose linkage configurations, are substantially less maneuverable and dexterous than biological analogs they ideally would emulate, notably the human arm or an elephant trunk. As previously stated, present jointed arm devices function much like backhoes, from a kinematic viewpoint, in that the arm linkage operates in a fixed plane which is rotated about one major vertical axis by the base revolute joint. With most of such devices, a given location and orientation of the tool corresponds to a single discrete set of joint angles and an associated discrete arm configuration. In a few of such devices, a given position and orientation of the tool can be achieved by two discrete arm configurations. An example of a device with two possible arm configurations for a single toolpoint position is the Unimation PUMA 600. In that device, while the revolute joints remain in a fixed plane, the "elbow" joint can be disposed either "up" or "down". Nevertheless, if for a prescribed position of the tool, an obstacle in the workspace interferes or the workpiece itself interferes with the arm segments, the arm is not capable of reaching the point without collision. Unlike the human arm, such conventional jointed arm manipulators do not have sufficient degrees of freedom to reach around the interfering object. This limitation is illustrated in FIG. 16. The human arm is considered to have seven degrees of freedom from shoulder to wrist, providing a range of elbow attitudes and resulting arm configurations for a given hand position and orientation. The elephant trunk, having more than seven degrees of freedom, can assume more complex configurations, and can "snake" between objects. Many automation tasks demand the dexterity of a human arm; some require even greater freedom of action. The lack of arm maneuverability and tool-handling dexterity in existing general-purpose computer-controlled manipulators presents serious limitations in their performance and adaptability to numerous applications.

The addition of one or more "redundant" joints in a manipulator has a number of significant benefits beyond improved maneuverability. In the same way that an extra degree of freedom provides means to reconfigure the arm to reach around an obstacle, the arm can be reconfigured to dispose joints in a way which distributes torque or velocity requirements among arm joints in the most equitable manner. A man reconfigures his arm in the process of lifting a heavy object to keep the forces and moments applied to each and every joint at a minimum. The man uses the redundancy in his arm to maximize "leverage". In a six degree of freedom jointed arm, operating in plane like a backhoe, no such reconfiguration and redistribution of forces and torques is possible. Thus, the mechanism's lifting capacity, associated with any particular point in its working envelop and discrete arm configuration, may be unreasonably limited because only a few joints are contributing to the exercise. With a kinematically-redundant manipulator, in contrast, while it may not be possible to lift a given load with one configuration of joints, it may be possible with another configuration and the arm can be reconfigured to do so. Similarly, in executing a high-speed move, the peak toolpoint velocity attainable by a six degree of freedom arm is ultimately determined at any one point in the path trajectory by the maximum speed of one joint. In a six degree of freedom arm, the motion requirements at any one point in the path again may not be well distributed among the joints, but no reconfiguration and redistribution is possible. The addition of redundant joints, therefore, promises to enhance greatly the efficiency of the manipulator, providing increased payload and applied tool force, as well as increased toolpoint speed, for a given amount of motive power and length of arm.

Another problem intrinsic to six degree of freedom arms which may be reduced by kinematic redundancy is related to joint travel limits. In the majority of mechanical embodiments of jointed arms, few, if any, of the revolute joints provide more than one full rotation. Many typical joints provide no more than 180 degrees of rotation. This feature limits the ability of the arm to accomplish certain motions. For example, if the prescribed path of movement specifies certain tool orientations, such as an orientation perpendicular in three axes to some arbitrary straight line in the workspace, then at some point in the straight line trajectory, one of the joints in the arm will reach its limit of travel and the desired path can be followed no further. This may be the case even when other joints remain close to their centers of travel. Kinematic redundancy provides a means to redistribute motion requirements in such a way as to maximize the use of all individual joint travel limits, thereby increasing the effective working envelop and tool-handling dexterity of the manipulator.

The implementation of kinematic redundancy in manipulator mechanisms can take many forms. Indeed, it can be achieved by the addition of one additional joint of any type at any location in any six degree of freedom arm linkage. Two colinear revolute joints in series will suffice. Such an additional degree of freedom need not be controlled in real time to achieve kinematic redundancy. A conventional manipulator providing six degrees of freedom under simultaneous control mounted on a slide that indexes the arm to different fixed positions during operation on the workpiece, offers a sort of primitive redundancy. However, in order to achieve human-arm-like dexterity from a conventional six degree of freedom general-purpose jointed arm manipulator, an additional revolute joint may be inserted between the shoulder "pitch" joint and the elbow "pitch" joint to allow the rotational axes of those two joints to move out of plane with respect to one another. This permits the elbow to be rotated out of plane, or "orbited", as shown in FIG. 18, providing the freedom to avoid obstacles and reach goalpoints on the back side of objects in the workspace, as shown in FIG. 17. In many accepted jointed arm designs, such as the ASEA IRb robots, the transmission linkages which are employed to transmit power to the remote arm joints make it difficult, if not impossible, to incorporate such a roll joint in the upper arm segment.

In order to achieve "intelligent", human-arm-like behavior with a kinematically-redundant arm, real-time sensory-interactive control is necessary. Such adaptive control of kinematic redundancy in manipulators demands that all of the seven or more joints be operated simultaneously and in concert by a real-time motion planning controller in response to information about internal arm conditions and to information from higher control levels and "off-board" sensors. The motion controller must handle both trajectory planning and coordinated joint control. It should reduce a programmed goalpoint to a set of coordinated joint commands, in real-time.

Conventional jointed arm robots also typically suffer certain significant limitations in performance related to the control stability and precision of movement of the manipulator. Many designs fail to provide servocontrol techniques which allow the high accuracy, repeatability and precision of movement required for applications such as metrology or assembly of small parts. The servocontrol systems of such manipulators may have limited operational bandwidth or may fail to employ important feedback control capabilities. As previously noted, some jointed arm designs incorporate mechanical features that further degrade stability and precision resulting from drive train compliance, structural compliance, and mechanical inaccuracies which are not effectively controlled by conventional machine tool servocontrol systems.

It is an object of the present invention to provide a manipulator with an arm geometry suitable for both general-purpose and a variety of special-purpose applications. It is a related object to provide an easily servicable modular manipulator design with standardized interchangeable arm segments which may be combined in various ways to construct manipulators of various sizes and load capacities. It is also a related object to provide a self-contained, durable manipulator capable of operating in crowded environments, harsh environments, and highly sanitized environments.

It is a further object of the present invention to provide a general-purpose manipulator with improved maneuverability, dexterity and repeatability. More specifically, it is an object to provide a manipulator and controller which may be configured to operate with seven or more axes, undeterred by the control problems which are introduced by kinematic redundancy. An allied object is to provide a manipulator with improved accuracy and repeatability.

In addition, it is also an object of the present invention to provide a manipulator and servocontrol which provide improved responsiveness of operation and smoothness of motion, enabling greater tool path speed and accuracy of movement.

Additional objects, advantages and novel features of the invention will be set forth in part in the description which follows, and in part will become apparent to those skilled in the art upon examination of the following, or may be learned by practice of the invention. The objects and advantages of the invention may be realized by means of the devices and methods described in the claims.

To achieve the above objects, the manipulator of this invention comprises an electronic controller and a set of unitized and integrated joint drive modules which can be assembled in series to form an arm. The manipulator is based upon an arm geometry design concept which enables the configuration of manipulators suitable to handle all general-purpose and special-purpose applications from three standardized joint types--roll, pitch, and transverse--assembled in different configurations. Each roll joint of the present invention consists of co-linear internal and external structural shells equipped for relative rotation by a reduction gearing assembly. Each pitch joint is disposed with its axis of rotation substantially perpendicular to and offset from the roll joint axes and consists of two relatively rotating parts, also driven by a reduction gearing assembly of essentially identical form to that employed in the roll joint. The present invention provides a design in which roll and pitch joints typically are joined in an alternating series to provide as many degrees of freedom and as much resulting maneuverability as is required by the particular application. The pitch joint axes of rotation are offset from the rotational axes of the roll joints. Each roll joint provides a minimum of 360 degrees of rotation. The offset and rotational range of each pitch joint permits adjacent roll joints to be folded back to be parallel to each other as shown in FIG. 19.

Each pitch joint may be provided with an ear module consisting of a body portion and a clevis arrangement of two parallel ears extending from the body portion. A case module, including a bell portion and a tubular portion extending from the bell portion, is provided such that the tubular portion may be mounted between the parallel ears of the ear module. A drive means also is provided for relative rotation between the case module and ear module and a resolver is provided at the maximum rotational radius to measure the relative rotation. A clutch is also provided to protect the gear reducer and actuator from damage in overload conditions.

With each roll joint there is provided an internal structural shell adapted to be secured readily to the adjacent case module of a pitch joint. There is also provided an external structural shell adapted to be secured to an adjacent ear module of a pitch joint. A drive means is provided to produce relative rotation between the internal and external structural shells of the roll joint and a resolver is provided to measure the relative rotation. An overload clutch is also provided. The above features permit the roll joint and pitch joints to be linked in a series to produce a manipulator with as many self-powered and easily interchangeable joints as are required for the application at hand.

A digital controller is also provided which guides the tool movement of the assembled manipulator along a path defined by a series of discrete positions designated by the user or, in the case of adaptive operation, by a host control system using sensors. A means is provided for storing the discrete positions in the data memory of the controller as cartesian coordinate values or for receiving such positions from a higher level real-time sensory interactive control. A means is provided for computing cartesian coordinate values of intermediate points between consecutive stored or received points, the intermediate points being spaced apart by equal time durations, so that the user-instructed routine is divided into a series of numerous short steps of movement of the tool relative to a given reference frame.

There also is provided a means for computing the angular displacement required from each pitch joint and roll joint of the manipulator associated with each of the intermediate points comprising the user-instructed path of the tool. A computing means for determining the joint angles is provided which will operate effectively for a manipulator consisting of any number of arm joints, including a redundant manipulator comprised of seven or more joints.

A servocontrol is provided for stablizing the motion of an apparatus such as the manipulator of the present invention. A servocontrol is provided which derives feedback from sensing the motive force on the actuator driven member which is responding to a velocity command signal. The motive force feedback is compared to the command signal and the resulting signal is compensated to provide stability and optimum bandwidth for the servomechanism. There also may be provided velocity and position feedback loops.

The present invention provides numerous advantages and benefits. The independently driven, unitized arm segment design permits the construction of a manipulator with any number of rotational axes suited to both general-purpose and special-purpose applications from a supply of standardized joint modules and components. This reduces the need for complete redesign when a customer requires a manipulator configured differently than those previously manufactured. Furthermore, the joint geometry permits the creation of a manipulator with unusual flexibility. The spherical work envelope provided by the manipulator is more complete than most manipulators, including those using the "three-roll wrist" which provides a sphere, less an approximately 90.degree. cone. Additional advantages also result from the unitized nature of the joints. The joints are provided with mounting features which permit a joint to be easily removed and replaced on an existing manipulator. This provides the ability to replace an inoperative joint quickly and easily. This ability may greatly reduce the down time of the manipulator because an inoperative joint can be replaced and the manipulator can be put back in operation while the troublesome joint is taken away for complete disassembly and repair. Furthermore, this feature permits large scale users of the manipulator of the present invention to stock relatively few standardized parts and joints for maintaining a large group of manipulators.

In this exoskeleton structure the principal flexural load bearing member in the roll joint arm segments is a thin-wall tube having a maximum suitable radius from the joint's axis of rotation. This provides an optimum stiffness to weight ratio in the link section. In both roll and pitch joints, the fully internal drives, transducers and wiring are protected from damage and the temperature of these components can be controlled with internal ventilation. Furthermore, the exoskeleton structure reduces the number of moving surfaces which must be sealed to one per roll joint and two per pitch joint. This helps to protect the internal drives, transducers and wiring from contamination, corrosive vapors and liquids. Indeed, with the selection of proper seals and with internal pressurization, the present invention can be made to be submersible in liquids. Conversely, in applications such as semiconductor manufacturing where the operating environment must be kept clean, the readily sealed exoskeleton structure may be used to keep machine-generated contaminants inside the machine.

The symmetrical bearing arrangement of the pitch joint modules provides maximum static and dynamic performance through balanced bearing and structural loading.

By mounting the resolvers at the maximum radius within each pitch joint and roll joint module, the highest possible ratio between joint and resolver armature can be achieved. Joint rotation is measured directly, in a thermally stable mounting, providing for high measurement accuracy and repeatability.

The relationship of dimensions of successive joints comprising the manipulator, including the offset of the pitch joint axes from the roll joint axes provides a manipulator of unusual maneuverability, dexterity and repeatability. The capability of the controller to control this manipulator in a sensory interactive manner, and indeed, to support kinematically redundant form, permits manipulator configurations, which can easily reach around objects and avoid obstacles as shown in FIG. 17. In addition, the offset of the pitch joint from the central axis of the machine, while presenting mathematical difficulties for the controller, enables the manipulator to fold back upon itself, thus making an arm of a given length and working envelope highly compact and contracted, thus increasing the effective working envelope.

The controller of the present invention also provides other significant operational advantages. As previously mentioned, the controller can effectively and efficiently guide the motion of a redundant manipulator. Moreover, the controller intelligently handles the computational difficulty presented by singularities. The controller also tends to avoid singularities in the work envelope of the manipulator.

Unlike typical machine tool designs, the servocontroller of a highly maneuverable manipulator such as that of the present invention requires special considerations to avoid sluggish and imprecise manipulator motion. The servocontrol of the present invention improves the smoothness and precision of motion of the manipulator by using feedback and compensation techniques to eliminate mechanical roughnesses in the manipulator system.

Additional advantages of the invention will be described in the description below and may become apparent to those skilled in the art upon reading the description or upon practicing the invention.

FIG. 1 is a perspective view of the manipulator according to the present invention with a diagrammatic illustration of the control system.

FIG. 2 is a side elevational view of the manipulator configured to include seven axes, illustrating the work envelope of the manipulator.

FIG. 3 is an exploded perspective view illustrating the means for assembling consecutive arm joints of the manipulator.

FIG. 4 is a semi-diagrammatic representation of the modularity feature provided by the present invention.

FIG. 5 is a top view of a portion of the manipulator arm extended with the rotational axes of adjacent roll joints colinear.

FIG. 6 is a side elevational view of the manipulator arm segment of FIG. 5.

FIG. 7 is a fragmented cross-sectional view of a roll joint taken along section line 6--6 of FIG. 5.

FIG. 8 is a fragmentary elevational view, partly in cross-section, of a pitch joint taken along section line 8--8 of FIG. 6.

FIG. 9 is a diagrammatic illustration of the relationship in an assembled manipulator between the cartesian coordinate values of the tool 16 relative to the manipulator base and the joint angle values of the pitch joints and roll joints.

FIG. 10 is a block diagram of the controller of the present invention.

FIG. 11 is a diagrammatic illustration of the computer hardware and circuitry boards comprising a portion of the controller represented in block diagram form in FIG. 10.

FIG. 12 is a graph used for illustrative purposes for explanation of the trajectory control means.

FIG. 13 is a block diagram of the circuitry required to interface the digital portion of the controller to the analog servocontrol circuits.

FIG. 14 is a block diagram of the system comprising the manipulator, the digital controller, and the servocontrol circuits.

FIG. 15 is a schematic diagram of the servocontrol circuits of the present invention.

FIG. 16 is an illustration of the maneuverability limitations of a typical jointed-arm manipulator.

FIG. 17 is an illustration of the maneuverability features of the present invention.

FIG. 18 is an illustration of the ability of the present invention to "orbit" the "elbow" of the manipulator out of a fixed plane of operation.

FIG. 19 is an illustration of the ability of adjacent roll joints of the present invention to fold back to a parallel position.

The present invention comprises an articulated anthropomorphic manipulator arm controlled electronically. A perspective view of a typical configuration of the present manipulator arm is shown in FIG. 1. The assembled structure consists of a linked series of arm segments or "joints" which extend from the base 1. The base 1 may be attached to a floor, overhead support structure, mobile track or other suitable support means.

The arm joints comprising the manipulator are of two basic types, roll joints, 2, 6, 10, and 14 and pitch joints, 4, 8, and 12. Except for variations in dimensions, a description of one pitch joint is representative of all pitch joints and a description of one roll joint is representative of all roll joints. That is, a manipulator arm could be built with any number of arm joints constructed according to the joint descriptions contained herein. It may be advantageous in some cases, however, to modify the design of the end joints 2 and 14 to accommodate specific applications or size limitations of components.

The pitch joints and the roll joints are independently powered to provide relative rotation about their longitudinal axes as shown by the arrows in FIG. 1. In a typical assembly such as that shown in FIG. 1, the first arm joint adjacent the base is roll joint 2. Each roll joint provides rotation about its longitudinal axis. The typical arm configuration includes a series of alternating pitch joints and roll joints terminating in an end effector or tool 16 at the distal end of the arm. As seen in FIG. 1, the roll joints typically decrease in cross-sectional area in the direction of the distal end of the arm. Similarly, the successive pitch joints following in that direction each typically reduce in size. On each typical pitch joint, such as pitch 8, the distal mounting collar 18 is smaller in diameter than the proximal mounting collar 20. While the arm joints of the manipulator reduce in size from the proximal to the distal end of the manipulator in the preferred embodiment, this feature is not a requirement of the present invention. By rotation of the various arm joints, the location and orientation of the tool 16 can be adjusted anywhere within the working envelope of the arm.

As illustrated in FIG. 1, and described below in greater detail, the arm joints are exoskeleton in nature. That is, the arm joints are comprised principally of tubular structures which provide the structural support necessary to carry a load at the tool location. Each arm joint is independently driven with its own internal motor and reduction gearing assembly contained within the exoskeleton structure. The exoskeleton structure provides the advantage of a high stiffness to weight ratio for the arm. The exoskeleton structure also protects the drive mechanisms as well as the wiring which, in the preferred embodiment, passes internally through the structure from joint to joint. In addition, the exoskeleton structure permits each of the drive motors to be air cooled, if necessary, by a single air source located in the base 1 of the arm. The exoskeleton design also may be sealed to permit underwater applications of the present invention or it may be positively ventilated to permit use in heavily contaminated environments.

FIG. 2 illustrates the working envelope of a manipulator arm according to the present invention. As seen from this illustration and others, each of the pitch joints, 4, 8, and 12 is offset from the longitudinal axis of the arm extended with roll joint axes colinear. This offset permits the manipulator arm to be folded back on itself or to be fully extended. This design feature provides a substantially spherical work envelope similar to that provided by a human arm. The outer line 19 represents the reach of the tool tip while the inner line 16 represents the reach when the final pitch joint 12 is folded back toward the roll joint 2.

FIG. 3 illustrates the manner in which successive arm joints are connected according to the present invention to build a manipulator. As explained below in greater detail, each roll joint and each pitch joint includes at both ends a mating face 30 and ramped ridge 32. In addition, the wiring necessary to accommodate the successive arm joints is carried internally and is connected to mating coupling assemblies 34 provided at each end of the arm joints. These features permit the arm to be easily constructed of successive arm joints, reducing in size toward the distal arm end.

To attach adjoining arm joints, the joints are brought into close proximity and the wire couplings 34 are engaged. Next the abutting faces 30 are placed in abutment. Finally, a clamp ring 36 is applied over the ramped ridges 32 of the adjoining arm joints and the ring 36 is tightened. In this manner, the successive arm joints are structurally linked and the wiring necessary to drive the more distal arm joints is connected. This clamping arrangement, described in greater detail below, provides the benefit of a uniform, continuous clamping force around the periphery of abutting segment faces 30 with as few as one bolt or other fastener. Consequently, successive arm segments can quickly be removed for repair and replaced with a substitute arm segment from inventory. Those skilled in the art will recognize that other fastening means between successive joints are possible. For example, each abutting end could be provided with an outwardly extending radial flange with a series of holes suitable to accommodate bolts or other appropriate fastening means though such a flange may limit the ability to fold the manipulator back upon itself.

The exoskeleton structure and independent drive means of each joint provides numerous benefits. As previously mentioned, an inoperative arm segment can quickly be removed for repair and replaced with an identical off-the-shelf arm segment. These features, together with the general reduction of size from proximal to distal arm ends, also permit the interchangeability of arm joints and internal drive assemblies across various size and strength classifications of manipulators as illustrated in FIG. 4. As seen in that figure, the pitch joint of a large manipulator nearest the distal end becomes the middle pitch joint of a smaller manipulator, etc. Similarly, the drive assembly of a roll joint near the distal end of a large manipulator becomes the drive assembly of a roll joint nearer the base of a smaller manipulator. Thus, a series of independently driven unitized pitch and roll joints can be produced which can be combined in various ways to provide a broad size and strength range of manipulators. More importantly, the number of successive arm joints combined to comprise a particular manipulator design can be varied widely. For example, for certain simple tasks it may be sufficient to have only three arm joints with a tool attached at the end of the third arm joint. Where greater maneuverability is required, the manipulator arm of the present invention can include any number of successive arm joints. As explained below, the preferred embodiment of the present invention consists of seven or more arm joints. By using seven or more arm joints, a manipulator can be constructed which will be able to reach around obstacles in various ways to apply a tool in a particular location and orientation. Furthermore, the use of seven or more arm joints creates a condition known as "redundancy" which provides numerous operational advantages if properly controlled.

Turning to FIG. 5, there is illustrated a top view of a portion of the manipulator arm extended with the roll joint axes colinear. The roll joints 6 and 10 have reducing cross-sectional areas in the direction of the distal end of the manipulator. The pitch joints 8 and 12 are positioned between each roll joint to provide rotation around their central axes 40 and 42, respectively.

The top view of pitch joint 8 in FIG. 5 provides some indication of the manner in which the pitch joints are constructed and the manner in which they operate. As previously stated, the joint includes a distal mounting collar 18 and a proximal mounting collar 20 which are of diameters suitable to accommodate the adjoining roll joints 10 and 6, respectively. Each pitch joint consists generally of two halves, an ear module 44 and a case module 46, which are joined by bearings and a drive means. The ear module 44 and the case module 46 pivot in relation to each other about the axis 40. A drive means for the pitch joint is contained within the pitch joint. The drive means extends longitudinally beyond the ears 48 and 50 of the ear module 44 and is enclosed by the motor cover 52 and the drive cover 54.

FIG. 6 is a side view of the structure of FIG. 5 and clearly illustrates that the rotational axis 40 of the pitch joint 8 is offset from the rotational axes 50 and 52 of the roll joints.

Turning now to FIG. 7, there is illustrated a partial cross-sectional view of the structure of FIG. 5 taken along section line 7--7. The pitch joint 8 is diagrammatically illustrated, including its axis 40. The roll joint 10 is made up of two basic parts, an inner drive housing assembly 64 and an outer skin 66. The inner drive housing and associated assembly 64 is rigidly secured to the case module 46 of pitch joint 8 by clamp ring 36. The outer skin 66 is mounted to the inner drive housing assembly 64 on bearings 68 and 70. When the motor 72 of inner drive housing assembly 64 is energized, the outer skin 66 rotates relative to inner drive housing assembly 64 and case module 46.

Turning to the drive housing assembly 64, there is provided an inner drive housing 65 with an outer bearing diagrammatically indicated at 70 mounted on an exterior bearing surface of drive housing 65. An annular seal 71 also is provided between drive housing 65 and outer skin 66 and a passage 73 is provided for passing wires. The inner race of the bearing 70 is clamped to the drive housing 65 by inner bearing clamp 74. The inner bearing clamp 74 is detachably fastened to the drive housing 65 by a series of bolts 76, or other suitable fastening means. The inner bearing clamp 74 includes an annular skirt 78 adapted to snugly fit within the annular recess provided in case module 46 of pitch joint 8. The inner bearing clamp on its outer periphery includes a sloped ridge 80. Likewise, case module 46 includes a sloped ridge 82 on its outer periphery adjacent the abutting surfaces between the case module 46 and the inner bearing clamp 74. The sloped ridges are adapted to accommodate a clamp ring 36. When the clamp ring 36 is applied and tightened, the inner bearing clamp 74 and the drive housing 65 become rigidly fixed to the case module 46.

Toward the distal end of the drive housing 65 there is provided, on its interior, two stepped annular surfaces 84 and 86 to accommodate the rigid spline 88 and the motor mount plate 90. The rigid spline 88 and motor mount plate 90 are rigidly affixed to the drive housing 65 by a series of bolts 92. The motor mount plate 90 includes a central aperature suitable to accommodate motor drive shaft 94. The motor mount plate 90 also is provided with a stepped recess 96 adapted to accommodate the collar of the motor 72. Motor mount plate 90 also includes an aperature through which suitable fasteners 98 can be applied to rigidly mount motor 72 to the motor mount plate 90. To the end of motor shaft 94 there is attached a wave generator 100 which is an integral part of the harmonic drive assembly. Wave generator 100 and the harmonic drive assembly are explained in greater detail below.

Near the distal end of drive housing 65 there is provided on the interior surface an annular interior seal 102. Likewise, there is provided on the exterior surface an annular bearing seat 104. The interior seal 102 maintains lubrication within the harmonic drive. The distal exterior bearing 68 maintains spacing and allows relative rotation between drive housing 65 and clutch ring 110.

Mounted within the drive housing near the distal end is a harmonic drive assembly. The harmonic drive assembly is a conventional commercially available item. The harmonic drive assembly includes flexspline 112 which is situated such that its exterior gear teeth are axially aligned with the teeth of rigid spline 88. The closed end of flexspline 112 is rigidly mounted via bolts 114 to harmonic drive collar 108. The harmonic drive collar 108 is provided with centering stubs 116 and 118. Centering stub 116 establishes the alignment of flexspline 112 so that flexspline 112 can be secured to harmonic drive collar 108 by bolts 114. Centering stub 118 performs a similar function on the distal side of the harmonic drive collar 108. Around centering stub 118 of harmonic drive collar 108 there is mounted a clutch plate 120 which is rigidly retained to the harmonic drive collar 108 by bolts 114. Harmonic drive collar 108 also includes a threaded central aperature 122 for receiving a sealing plug to retain lubricants and provide access to adjust the wave generator 100.

Clutch plate 120 extends in the distal direction and flares outwardly radially to provide an annular sloped ridge 124 and a radial abutment surface 126. Clutch ring 110 also provides a distal region with a corresponding annular sloped ridge 130 and a cooperating radial abutment surface. When the abutment surfaces 126 and 127 of clutch plate 120 and clutch ring 110 are brought into abutment and a ring clamp 128 is attached to clutch plate 120 and clutch ring 110 engaging sloped ridges 124 and 130, this assembly provides a clutch for transmitting the rotary motion from the motor 72 and harmonic drive assembly to the outer skin 66. This is explained in greater detail below. Clutch ring 110 has mounted on surface 129 several strain gauges which are used for feedback control of the arm motion.

Towards its proximal end, clutch ring 110 provides an annular bearing seat for the outer race of the distal exterior bearing 68. This race is retained on the clutch ring 110 bearing seat by an annular bearing clamp 132 and a plurality of bolts 134 or other suitable fasteners. Clutch ring 110 also includes arcuate aperatures, two of which are shown at 136. These aperatures provide a space for the internal wiring to pass from joint to joint. The outer circumference of clutch ring 110 includes an annular bolt rim 138. The bolt rim 138 contains a number of aperatures suitable to accommodate a series of bolts 140. The bolts 140 pass through clutch ring 110 to threadedly engage internal annular lip 142 of the distal end of outer skin 66 and engages flange 67 of the proximal end of distal outer skin 69.

The outer shell of the roll joint may consist of a single outer skin 66 or may include a distal outer skin 69. In the preferred embodiment, the transverse cross-section of the outer skin 66 and distal outer skin 69 increase in the direction of the proximal end of roll joint 6. At the proximal end on the interior diameter there is provided a bearing seat 144. The outer race of bearing 70 is retained on the bearing seat 144 by an annular outer bearing clamp gear 146 which is rigidly attached to the outer skin 66 by a series of bolts 148. Outer bearing clamp gear 146 is provided with internal gear teeth 150. The internal gear teeth 150 are adapted to mesh with anti-backlash gear 152 which protrudes through arcuate aperatures (not shown) in the inner bearing clamp 74 and drive housing 65. As an additional function, outer bearing clamp gear 146 provides a wear surface for a tube seal 154. Tube seal 154 is held in place by a recess provided in the distal face of inner bearing clamp 74.

Inner bearing clamp 74, in addition to the features described above, is provided with a resolver mounting surface 156. The resolver mounting surface 156 contains an aperature adapted to accommodate a mounting flange and pilot 158 of a standard, commercially available resolver 160. The resolver 160 is retained within the mounting aperature by clamp 162. The anti-backlash gear 152 is mounted to the resolver shaft by clamp 164. With this mounting, the resolver is able to provide a measure of relative rotational position between the drive housing 65 and the outer skins 66 and 69.

In the preferred embodiment, motor 72 includes both a brake and, optionally, a tachometer. In operation, an electrical current is supplied to motor 72 causing the wave generator 100 of the harmonic drive assembly to rotate within the flexspline 112. This action induces a counter-rotating motion in the flexspline 112 which is transmitted to the rigidly attached harmonic drive collar 108. The rotating motion is transmitted through the collar 108 to clutch plate 120. The clutch assembly guards against overloading of the harmonic drive assembly. The clamping force of ring clamp 128 is set to allow slippage between clutch plate 120 and clutch ring 110 in the event the torque load on the roll joint exceeds the torque capabilities of the harmonic drive assembly. If the torque load does not exceed this limit, the clutch assembly transmits the rotational movement of the harmonic drive assembly to the clutch ring 110 and ultimately to outer skin 66. As previously described, this relative rotation between the drive housing assembly 64 and the outer skin 66 assembly is accommodated by outer bearing 70 and distal exterior bearing 68. As the outer skin 66 rotates relative to the drive housing assembly 64, the outer bearing gear clamp 146 moves in relation to the resolver 160 location. The gear mesh between outer bearing gear clamp 146 and anti-backlash gear 152 causes rotation of the resolver shaft 166. In this manner, the resolver 160 can be used to measure rotation of the roll joint 10 and relative rotational position of the outer skin 66 with respect to the drive housing 65.

At the distal end of distal outer skin 69, there is provided an annular abutting face 167, sloped ridge 168, and interior annular lip 172. Ear module 170 of pitch joint 12, as shown in FIG. 5, includes an identical annular abutting face 169 and a corresponding sloped ridge 171. The inner diameter of the ear module 170 is adapted to closely fit on the lip 172. When the ear module 170 and outer skin 66 are mated and the distal ring clamp 174 is applied, the ear module 170, ring clamp 174, and distal outer skin 69 are rigidly joined.

Turning now to FIG. 8, there is illustrated a partial cross-sectional view of the pitch joint 8. As with a roll joint 10, the pitch joint 8 includes two major structural parts which rotate relative to each other. These relatively rotating parts are the ear module 44 and case module 46, also visible in FIG. 4. As shown in FIGS. 3, 5, and 6, the ear module 44 of each pitch joint is disposed in the manipulator arm assembly nearest the proximal end of the manipulator arm while the case module 46 is disposed nearest the distal end of the manipulator arm. Thus, for simplicity it is sensible to view the pitch joint as operating with a fixed ear module 44 and a rotating case module 46.

As the name would suggest, the ear module 44 includes two parallel ears, motor ear 176 and drive ear 178 extending outwardly toward the distal side of the pitch joint. Ear module 44 also includes in its bell-shaped portion an aperature 45 for passing wires from the roll joint 6 into the annular cavity of the pitch joint surrounding motor 256. The ears 176 and 178 are disposed in parallel planes which are parallel to the rotational axes of the adjoining roll joints 6 and 10. (See also FIGS. 5 and 6.) Each ear 176 and 178 includes an aperature suitable for accommodating a bearing inner race support assembly.

Turning first to the motor side of the pitch joint (the lower portion of FIG. 8), the motor ear 176 bearing support assembly is shown in detail. The motor ear 176 is provided with an aperature suitable to accommodate a snugly fit inner bearing clamp 180. The motor side inner bearing clamp 180 contains a radially extending flange 182 which abuts the outer surface of the motor ear 176. Spaced aperatures are provided in the flange 182 to allow passage of bolts 184 which are threaded into mating threaded holes around the perimeter of the motor ear 176 aperature. The motor side inner bearing clamp 180 also includes a ring of aperatures for accommodating bearing bolts 186. These bolts draw together the motor side inner bearing clamp 180 and the motor side inner bearing collar 188. Inner bearing collar 188 is provided with an annular bearing seat 190. The inner race of motor side bearing 192 is seated upon the bearing seat 190 and clamped in place by the bolt force holding together the motor side inner bearing collar 188 and the motor side inner bearing clamp 180. The inside face of the motor side inner bearing collar 188 includes a stepped surface suitable to accommodate an external tooth gear 194. The external tooth gear 194 is retained on the motor side inner bearing collar 188 by bolts or other suitable fastening means (not shown).

Turning to the drive side, a similar bearing mounting arrangement is illustrated. The drive side ear 178 includes an aperature suitable to accommodate the intermediate surface of inner bearing clamp/clutch 196. Inner bearing clamp/clutch 196 includes a flange 198 with spaced aperatures suitable to accommodate a number of bolts 200 which engage threaded holes around the periphery of the central aperature in drive ear 178. The inner bearing clamp/clutch 196 is also provided with an internal array of aperatures suitable to accommodate bolts 202 which engage threaded holes in the mating drive side inner bearing collar 204. The drive side inner bearing collar 204 includes on its outer surface a bearing seat 205. The inner race of bearing 206 is mounted on the bearing seat 205 and retained between the drive side inner bearing collar 204 and the inner bearing clamp/clutch 196 by the clamping force of bolts 202. In addition, the drive side inner bearing collar 204 includes a mating surface 207 on its outer diameter near the inner end of the collar. This surface is suited to accommodate an annular seal 208. The end of the inner bearing clamp/clutch 196 includes a flat abutment face 197 and a sloped ridge 210. The flat annular abutment face 197 is adapted to mate with an identical abutment face 215 and a sloped ridge 212 on the harmonic drive clutch plate 214. A ring clamp 216 is secured over the ramped ridges 210 and 212 to provide a clutch mechanism similar to that found in the roll joint of FIG. 7. Surface 211 of inner bearing clamp/clutch 196 includes several strain gauges which are used for feedback control of the arm motion.

In addition to the sloped ridge 212, the harmonic drive clutch plate 214 includes a centering stub 218 the outer diameter of which roughly mates with the adjacent inner diameter of the inner bearing clamp/clutch 196. The harmonic drive clutch plate 214 includes a circular array of aperatures which accommodate bolts 220. Clutch plate 214 also includes a central aperature 222 suitable for a plug to retain lubricant and permit access to the wave generator 260 for adjustment. Bolts 220 extend through clutch plate 214 to engage threaded holes in harmonic drive collar 224. The harmonic drive collar 224 includes a center stub which mates a central recess 217 in the harmonic drive clutch plate 214. These control surfaces ensure centering of the various parts. The harmonic drive clutch plate 214 and harmonic drive collar 224 mate and retain between their surfaces the flexspline 226 of the harmonic drive assembly.

Turning now to the case module 46, there is provided a generally tubular body slightly shorter than the distance between the ears 176 and 178 of the ear module 44. At each end of the tubular case module 46, on the inner diameter, there is provided annular bearing seats 47 and 51 for the outer races of bearings 192 and 206, respectively. On the drive side of the case module 46 inner diameter, there is further provided an annular seal contact surface 49 for seal 208. The outer races of the motor side bearing 192 and the drive side bearing 206 each are retained on their respective bearing seats by the motor side outer bearing clamp 228 and the drive side outer bearing clamp 230, respectively. The case module also includes a slotted region 232 opposite the ear module 44 which accommodates the passage of electrical wires.

The case model 46 includes an integral radial flange 234 which extends inward to provide mounting surfaces. At the center of this flange there is provided an aperature which is concentric with the axis of rotation 40 of the pitch joint 8. The aperature is adapted to accommodate a motor mount plate 238. The motor mount plate 238 includes an annular array of aperatures suitable to accommodate bolts 240 which engage threaded holes in the case module interior flange 234. The case module interior flange 234 also provides on its drive side a surface 242. The rigid spline 244 of the harmonic drive assembly is rigidly mounted to the surface 242 by bolts 246 which extend through the gear to engage threaded holes in the case module interior flange 234.

The motor mount plate 238 includes a central aperature suitable to accommodate passage of the motor drive shaft 248 and a suitable seal 250. The motor mount plate 238 includes a circular recess 252 on its motor side which is provided to accommodate the centering collar 254 of the motor 256. The motor mount plate 238 also provides recessed aperatures through which appropriate screws or bolts 258 can be passed to rigidly mount the drive motor 256 to the motor mount plate 238. Mounted on the motor shaft 248 is a wave generator 260, a part of the harmonic drive assembly.

In operation, an electric current is supplied to motor 256 causing rotation of the motor drive shaft 248 and of the wave generator 260 of the harmonic drive assembly. This creates relative movement between the rigid spline 244 and the flexspline 226 of the harmonic drive. Viewing the ear module 44 as fixed, the result is that the rigid spline 244 rotates around the flexspline 226 inducing rotation of the case module 46 relative to the ear module 44. Bearings 192 and 206 are interposed between the ear module 44 and the case module 46 to accommodate the rotation. If the torque load encountered by the pitch joint exceeds the frictional resistance provided by the clutch assembly, the case module 46 and ear module 44 will remain fixed in their relative location and the harmonic drive clutch plate 214 will rotate relative to the inner bearing clamp/clutch 196. Thus, when the clutch is properly set it will prevent damage to the harmonic drive assembly.

When relative rotation occurs between the ear module 44 and the case module 46, rotation of the shaft of resolver 259 results. This occurs because the resolver 259 is mounted in the case module 46 while the teeth of the anti-backlash gear mounted on the resolver shaft (not shown) engage the external tooth spur gear 194 which is rigidly mounted to the ear module 44. This arrangement provides a means for detecting the relative rotation and the relative rotational position between the case module 46 and ear module 44.

As previously explained, the roll joints and pitch joints described above can be joined in alternating sequence to construct a manipulator arm with any number of arm joints. Those skilled in the art will recognize that a wide variety of arm configurations are possible with the present invention. In addition to the conventional configuration of alternating pitch and roll joints, for example, a manipulator could be constructed from pitch and roll joints of the present invention in which two or more adjoining joints are of the same type, i.e. pitch or roll. Also, a simple yaw joint could be placed between two pitch or roll joints to provide motion about a vertical axis. It would also be apparent to those skilled in the art that a "dead" static joint of any shape could be substituted in the arm. Similarly, the perpendicular relationship between roll joint and pitch joint axes of rotation provided in the preferred embodiment could be altered. This wide variety of configurations is made possible and simple by the exoskeleton design of the arm segments and by the fact that each arm joint is independently powered.

Those skilled in the art will recognize that the unitized nature of the arm joints facilitates the construction of a redundant manipulator which is advantageous in operation but also problematic. Redundancy generally improves the maneuverability of the manipulator. However, to be useful, the manipulator motion must be accurately controlled and redundancy makes that task more difficult. The controller of the present invention is uniquely adapted to deal with redundancy and with other motion attributes of a multi-axis jointed manipulator arm. The controller of the present invention therefore permits operability of the wide variety of arm configurations made possible by the unitized arm segments described above.

FIG. 1 includes a block representation of the control system 3 of the present invention. The control system 3 is an integral part of an operational manipulator arm. Without a control system, a manipulator arm has little utility.

As represented by FIG. 1, the control system 3 can be viewed for simplicity of explanation as having several discrete elements. The control system 3 of the present invention includes an input/output console 5 to permit communication with an operator or to accommodate various process or machine inputs. It also includes a digital computer 11 which processes a user-program contained in the data memory 9 in accordance with the control programs contained in program memory 7. The user-program directs the actions of the manipulator. The control system 3 also includes analog servocontrol circuits 13 which provide communication with the arm joints and tool 16. Numerous additional peripheral devices may be included to interface the manipulator to the control system 3. For purposes of this discussion, the program memory 7, data memory 9, digital computer 11, and analog servocontrol circuits 13 will be collectively referred to as the controller 15, shown within the phantom box.

The operating system of the controller 15 includes a mode control program. This program allows for switching the controller 15 between various operating modes such as automatic mode, teach mode, manual mode, etc. During the teach mode, an operator uses input/output console 5 to actuate the drive motors of the various arm joints and thereby move the tool 16 into desired locations and orientations along a desired path (i.e. "Goal Positions"). By so doing, the operator establishes in data memory 9 numerous "Goal Positions" on a path of movement which the manipulator will later follow. For example, if the operator wishes for the manipulator to execute a series of spot welds on an object then return to a home position, the operator will run the manipulator through that routine and cause position data to be stored in data memory 9 only when crucial positions are reached, such as the spot weld position. This exercise establishes Goal Positions defining a routine which the manipulator later will follow. For purposes of this description, the term "position" includes both the three-dimensional location of the tool 16 and the orientation of the tool 16.

Additional functions may also be provided with the teach mode. While numerous other operating modes may be available, only the automatic mode will be described in detail here since that is the principal mode involved in the use of the present invention. Operation of the automatic mode assumes that the operator already has "taught" the manipulator the Goal Positions of the routine it is to carry out and that the appropriate information required for execution of the automatic mode is stored in data memory 9 of the controller 15.

The objective of the automatic mode of the controller 15 is to precisely control the location and orientation of tool 16 in space. Referring to FIG. 1, the objective of the controller 15 is to guide the motion of the manipulator so as to precisely control the position of tool 16. This is accomplished by providing a rapid succession of digital signals, each a result of a complex series of computations, which are converted to analog signals and transmitted to each arm joint via analog servocontrol circuits 13. Using appropriate feedback and compensation techniques, the servocontrol circuits 13 provide a final continuous drive signal to each joint motor.

The location of the end point of tool 16 in space can be specified with respect to the cartesian coordinate system shown at the manipulator base in FIGS. 1 and 9. That is, the location of the end point of tool 16 can be uniquely defined by values of X, Y and Z in the cartesian coordinate system. The orientation of tool 16 can be defined by reference to a localized cartesian coordinate system whose origin is the end of tool 16 or at the end of the last joint. The orientation can be represented in this local cartesian coordinate system by three values--i.sub.z, j.sub.z, k.sub.z --which equal the length of the projections of a vector of one unit length on the X, Y and Z axes of the localized cartesian coordinate system in which the local Z axis is coincident with the tool centerline. If rolling motion of the tool 16 around its own axis also needs to be controlled, a supplemental set of i.sub.x, j.sub.x, and k.sub.x values corresponding to a local axis perpendicular to the local Z axis can be added. In general, however, any location and orientation (both direction and roll) of the tool 16 can be satisfactorily defined by a set of six values--the X, Y and Z locations of the tool end point relative to the universal cartesian coordinate system and the i.sub.z, j.sub.z, and i.sub.x values the tool 16 defined by the local cartesian coordinate system.

FIG. 9 illustrates the geometric relationships involved in the control of a jointed manipulator without the joint offsets present in the illustration. The figure shows the axes of the roll joints of the manipulator of FIG. 1 placed in a cartesian coordinate system. To enable the manipulator to readily adapt to changing Goal Positions or obstacles in its environment it is convenient to define the tool end point 16 location and tool orientation in cartesian coordinates values, X, Y and Z and i.sub.z, j.sub.z and i.sub.x. However, movement of the tool position is accomplished by changing the roll and pitch joint angles .theta..sub.1 through .theta..sub.7. The relationships between the tool position and the joint angles of the seven-axis manipulator of the preferred embodiment are defined by the transformation equations below. Similar transformation equations exist to define the geometric relationships involved with manipulators with other numbers of axes or joints. Note that the offset of the pitch joints complicate the formulas and that the formulas are unique to the geometry of the particular machine.

The orientation equation, relating tool orientation to joint angles: ##EQU1## where [.theta..sub.i ] is the rotational matrix for joint i given by the following equations: ##EQU2##

The location equations, relating tool location to joint angles: ##EQU3## where: R.sub.i is the distance by which the rotational axis of pitch joint i is offset from the rotational axes of the adjacent roll joints;

L.sub.i is the distance between rotational axes of the two pitch joints adjacent roll joint i; and ##EQU4## represents the distance by which the tool is offset from the end of the last distal manipulator joint.

Movement of the tool 16 in space from one location and orientation to another is accomplished in the manipulator described above by energizing the motors to cause relative rotation within the roll joints and pitch joints. To accomplish movement of the tool 16 between successive Goal Positions stored in data memory 9 as cartesian coordinate values, therefore, it is necessary to determine the appropriate direction and magnitude of rotation of each roll joint and pitch joint. As previously mentioned, there are numerous advantages that arise from the use of manipulators which use rotational joints to effect displacement and orientation of a tool. However, such revolute or anthropomorphic manipulators are difficult to control since the arm joint rotations cannot easily be mathematically correlated to the three-dimensional movements stored in cartesian coordinate values. In other words, it is difficult to mathematically compute quickly the amount by which each of the arm roll and pitch joints should be rotated in order to accomplish a movement in space of the tool tip between successive Goal Positions stored as cartesian coordinate values in data memory 9. It will be readily apparent to anyone skilled in the art that, even in the absence of the kinematic redundancy provided with a manipulator of more than six axes, it is difficult to obtain precise numerical solutions for the joint angle values from the above transformation equations in a timely manner using conventional digital computer technology.

In addition to the difficulty of rapid joint angle computation described above, the seven-axis manipulator of the preferred embodiment presents further control problems. The number of joint angles which must be determined to uniquely define a specified tool position described in cartesian coordinates equals the number of arm joints or axes comprising the manipulator arm. A problem arises when the manipulator arm consists of more arm joints than there are known cartesian coordinate values. For example, if the location and orientation of the tool 16 is completely specified with cartesian coordinate values of X, Y, Z and i.sub.z, j.sub.z, i.sub.x, a manipulator arm consisting of seven joints, as is the case in the preferred embodiment, will yield six equations with seven unknown joint angles which must be calculated to direct the movement of the tool 16. This condition provides kinematic redundancy--advantageous for maneuverability and obstacle avoidance but difficult to control for effective operation.

The controller 15 of the present invention deals effectively with the kinematic redundancy provided when the manipulator is constructed of seven or more joints. It also enables rapid, efficient computation of joint angle values as prescribed by the transformation equations above. Furthermore, the controller 15 of the present invention is able to deal effectively with a condition which arises in the control of both redundant and non-redundant anthropomorphic manipulators known as a singularity. This condition, and its treatment, is described below.

As previously stated, the controller 15 in operation provides a rapid succession of digital signals which eventually result in drive signals to each joint motor. In general, the digital processing carried out in controller 15 operates in the following manner when in the automatic mode. Referring to FIG. 10, the Move Control 270 of controller 15 sequentially retrieves the user-instructed operations and Goal Positions for the tool 16 stored as cartesian coordinates. This and related information are supplied to the Trajectory Control 272 which computes the cartesian coordinates of intermediate points along the instructed path spaced 33 milli-seconds apart--the cycle time of the first-tier digital circuitry--(hereinafter "Intermediate Points"). This and related information is supplied to the Redundant Coordinated Joint Control 274 (hereinafter "RCJC") which transforms the set of cartesian coordinate values defining the next Intermediate Point into corresponding joint angle values. The RCJC is able to perform such a computation for a manipulator of any number of joints. These values are transmitted to the Servo Control Interface 276 which adjusts the values for environmental conditions or manipulator physical limits, if necessary. The Servo Control 278 then receives the adjusted joint angle values and generates a digital position error signal for each joint which is proportional to the difference between the actual joint angle value at that instant in time and the value required at the end of the next 3 milli-seconds--the cycle time of the second-tier digital circuitry--to achieve the next Intermediate Point at the appointed time. These digital position error signals are converted to analog signals by the Resolver Boards 280 and transmitted to the Analog Servocontroller Circuits 282 where appropriate feedback and signal compensation is performed, yielding a drive signal for each joint motor. This digital processing is described more fully below and more fully disclosed in the program listings in the Appendix. These programs also could be embodied in a proprietary read-only-memory chip ("ROM").

In the preferred embodiment of the present invention, the digital processing is performed on the hardware and in the manner diagrammatically illustrated in FIG. 11. An Intel 86/30 circuit board designed by the Intel Corporation with an 8086/8087 co-processor and 128K of dual port memory is run on a 33 milli-second clock cycle in tandem with an Intel 86/05 circuit board with an 8086/8087 co-processor and 16K of static non-dual port memory. The user instructed Goal Position and goal path data 284 are stored on the 86/30 circuit board. The 86/30 circuit board also contains in memory the Move Control program 286 and a portion of the RCJC program 290 which formulates part of a Jacobian matrix.

The 86/05 circuit board contains in memory the Trajectory Control program 288, the remaining portions of the RCJC program 290 and all of the Servo Control Interface program 292. The 86/30 and the 86/05 are connected to communicate by a designated protocol on a multibus 300. Also connected to the multibus 300 are the Resolver Boards 296 and 298 described above and an Intel 86/05 circuit operating on a 3 milli-second clock cycle. This 3 milli-second 86/05 contains in memory the Servo Board program 294.

The digital data processing among these components occurs in the following sequence The Move Control program 286 processes the user-instructed data 284 on the Intel 86/30 to compute the next Goal Position and related data. This data is transmitted over the multibus 300 to the Intel 86/05 operating on the 33 millisecond clock cycle. There the data is processed by the Trajectory Control program 288 to provide Intermediate Point cartesian coordinate values. These Intermediate Point values and related data are sent to the 86/30 processor where a portion of the RCJC program 290 processes them to formulate part of a Jacobian matrix necessary for the redundant joint algorithm described below. The other portion of the RCJC program 290 on the 33 milli-second 86/05 simultaneously formulates the remaining portion of the Jacobian matrix. The portion of the Jacobian matrix formulated by the 86/30 processor is transferred via the multibus 300 to the 33 milli-second 86/05 where it is combined with the other portion of the Jacobian matrix and used in the RCJC program 290 to determine the Intermediate Point joint angle values in accordance with the redundant joint algorithm described below. The Intermediate Point joint angle values then are processed in the 33 milli-second 86/05 using the Servo Control Interface program 292 to compute the corresponding resolver positions for each arm segment and the time allocated to attain those resolver positions. Normally, the time allocated will be 33 milliseconds--the time set for cycling of the Intermediate Point joint angle recomputation. However, a longer time period may be allocated if it is necessary to achieve greater joint angle accuracy through additional iterations of the RCJC program 290 or if it is necessary to scale back the time to adjust for motion commands which are beyond the physical limits of the arm. The resolver positions and allocated time are transmitted over the multibus 300 to the 3 milli-second 86/05. The 3 milli-second 86/05 compares that data to the true current resolver position data most recently received over the multibus 300 from the resolvers via the Resolver Boards 296 and 298. The 3 milli-second 86/05 processes these signals as provided by the Servo Board program 294 and generates digital position error values for each of the arm joints every three milli-seconds. The digital position error values for each arm joint are transmitted over the multibus 300 to the appropriate Resolver Board 296 or 298. There the digital signals are converted to analog signals and further processed in the Analog Servocontroller Boards 282 as explained below to provide a final drive signal to each joint.

A description in greater detail of the operation and various features of the programs 286 through 294 is presented below. In addition, a complete enabling disclosure can be found in the Appendix where a listing of each of the programs 288 through 294 indicated in FIG. 11 is presented.

The Move Control 270 operates primarily as an interface between the Trajectory Control 272 and the instructions and Goal Position data supplied by the user. Referred to FIG. 10, the Move Control 270 sequentially retrieves and interprets the instructions stored by the user in data memory 9 in order to identify the Goal Position that is next in time. In addition, the Move Control 270 supplies the user-instructed intermediate velocity to be maintained by the tool 16 during movement toward the next Goal Position and supplies the user-instructed final velocity to be achieved upon reaching the Goal Position. Furthermore, the Move Control 270 specifies a type of approach path to be followed, such as straight line or joint space. For purposes of this description, the collection of data defining the movement of tool 16 between two consecutive user-instructed Goal Positions will be referred to as the "Move Data" and a "Move" will refer to the interval of movement defined by the two consecutive user-instructed Goal Points and corresponding velocities and path definitions.

The Move Data computed by the Move Control 270 define the next Move to be made by the manipulator arm. This data periodically is received by the Trajectory Control 272. At each interval of computation by the controller 15, (the first-tier of the controller 15 operates on a 33 milli-second clock cycle) the most recent known tool 16 position, defined in global cartesian coordinates (X, Y, Z) and local orientation values (i.sub.z, j.sub.z, i.sub.x), and the most recent known velocity are carried in the memory of the Trajectory Control 272. Using the current position and velocity information and the Move Data for the next Move provided by the Move Control 270, the Trajectory Control 272 calculates the coordinate values in cartesian coordinates which the tool 16 must attain at the end of the next successive 33 milli-second time interval in order to reach the user-instructed Goal Position in accordance with the Move Data and pre-programmed joint acceleration rates. In this manner, a Move between successive user-instructed Goal Points A and B, for example, is divided into the numerous "Intermediate Points" spaced 33 milli-seconds apart.

An example of the computations executed by the Trajectory Control to compute the Intermediate Points of each Move may be described in general terms. For simplicity of illustration, the computations will be described in connection with a typical Move. The Move used for illustrative purposes is a Move of the tool end point 16 from point A to point B with a zero velocity at the beginning of the Move (i.e. at point A), a zero velocity at the end of the Move (i.e. at point B) and an intermediate velocity of V.sub.I. As mentioned above, the movement between any two successive user-instructed Goal Positions occurs in a finite number of small intermediate steps spaced 33 milli-seconds apart. Each of the Intermediate Points defining these intermediate movements are recalculated by the controller 15 at each 33 milli-second interval. Thus, a Move consists of numerous Intermediate Points along a particular path which are computed only milliseconds before they actually are reached by the tool 16.

To be able to compute the Intermediate Points during the Move, the Trajectory Control 272 first calculates certain parameters associated with the velocity profile of the Move. In the example shown in FIG. 12, the graph illustrates the velocity profile of the illustrative move from A to B. The initial velocity, V.sub.o, and final velocity, V.sub.f, are zero, and the intermediate velocity equals V.sub.i. At the beginning of the Move, the values of V.sub.o and V.sub.f, are known. By counting the passage of time from the beginning of the Move and correlating the elapsed time with the velocity profile of the Move, the distance to the next Intermediate Point can be computed.

The trajectory method can be understood by reference to FIG. 12. The distance of any given Intermediate Point along the Move path from the preceding Intermediate Point, to be attained 33 milli-seconds later, depends upon the location of the tool tip on the velocity profile of FIG. 12 at that instant in time. For example, if we look at the first 33 milli-second interval of movement from the starting point A, the next intermediate tool tip location will be a very short distance from point A because the arm at that instant in time will be moving at a low velocity. However, during the middle of the Move from A to B, the tool tip is commanded to move at a velocity of V.sub.i. During that segment of the Move, the distance between any two consecutive 33 milli-second Intermediate Points will be much greater. In either case, the distance to the next consecutive Intermediate Point from the starting point A is proportional to the percentage of area subsumed under the velocity curve at that instant in time. This technique is independent of the path type.

The Trajectory Control 272, therefore, first calculates certain key parameters associated with the velocity profile of the particular Move type about to be executed. These Move parameters are: t.sub.a.sup.0, t.sub.a.sup.1, t.sub.final, d.sub.a.sup.0k, and d.sub.a.sup.1. It will be apparent to one skilled in the art from viewing FIG. 12 and the formulas below how these Move parameters may be calculated. ##EQU5## Once these critical time points of the velocity profile are calculated they are used in the Trajectory Control 272 to compute the Intermediate Points of the Move. The detailed operation of the Trajectory Control program can be seen from the listing in the Appendix.

The Trajectory Control program also discloses a novel method of tool orientation. As previously stated, successive goal positions are defined by location values and orientation values. The tool orientation method of the present invention accomplishes reorientation between successive goal positions in a smooth continuous motion, following a geodesic path. The method entails (i) finding a vector about which the tool orientation triad of the current position can be rotated to attain the tool orientation triad of the next successive goal position and (ii) finding the magnitude and direction of rotation required about that axis. These items of information are computed using an Eigenvector technique in which the axis of rotation is given by the Eigenvector and the angle of rotation is given by the Eigenvalue. Using this information, the tool is re-oriented between successive goal positions in a smooth, continuous motion at a rate consistent with the motion along the path between successive goal positions.

Refering again to FIG. 10, the cartesian coordinate values of the next Intermediate Point are transmitted from the Trajectory Control 272 to the Redundant Coordinated Joint Control 274 ("RCJC"). The RCJC 274 contains in memory the angular values of the arm joints, ("joint angles") associated with the most recent Intermediate Point. The RCJC also has stored in memory certain "modal parameters" associated with the given Move. Modal parameters are conditions defined independently of the operator-instructed Goal Points and path type, but which ultimately affect the coordination of movement of the arm. Modal parameters include the tool 16 offset location and direction; the linear, orientation, and arm segment acceleration rates; the accuracy tolerance; the tool vector roll constraint (on or off); and the joint spring stiffnesses and spring origins. Some of these items are used by the Trajectory Control 272 and others are used by the RCJC 274.

Given these modal parameters and the most recent known joint angle values, the RCJC 274 calculates the joint angle values which must be achieved to reach the cartesian coordinate values determined for the next Intermediate Point by the Trajectory Control 272. To understand the method by which the RCJC computes these Intermediate Point joint angles, some background information about manipulator motion and associated mathematics is required.

As previously stated, the preferred embodiment of the present invention consists of a manipulator with seven arm joints--i.e. a kinematically redundant manipulator. This presents control difficulties because the transformation equations which relate cartesian coordinate values to joint angles have seven unknown variables (joint angles .theta..sub.1 through .theta..sub.7) but only six known variables (cartesian coordinate values of the tool 16 endpoint). If such kinematic redundancy is present, it can be handled by specifying an extra mathematical criteria to compute the joint angles. In the preferred embodiment of the present invention, an extra criteria (an "objective function") is used to control redundancy. This will be explained more fully below.

Another problem which must be considered in controlling the motion of a jointed, multi-axis manipulator arm such as that shown in FIG. 1 is a condition known as "singularity". Singularity is the condition of the loss of one or more degrees of freedom of movement in one direction and redundancy in another direction. Sets of singularities occur only at unique joint angles. Its occurrence depends upon the geometry of the arm (i.e. the dimensions of manipulator); the position of the arm at that instant in time; and the direction of movement the arm is called upon to make at that instant; and other possible factors.

The loss of mobility which is a property of a singularity is undesirable and should be avoided in the movement of the manipulator arm. Numerous joint configurations will give rise to a singularity regardless of the number of arm joints of which the manipulator is comprised. With a multi-axis arm such as that described above moving in three dimensional space, many singularity points exist.

When a manipulator arm approaches a set of singularities, one or more of the joints of the machine is effectively eliminated from contributing to the movement the arm is directed to execute. Thus, when a singularity is approached, a six-axis machine effectively is reduced to five degrees of freedom of movement, a five-axis machine to four, etc. In addition, as a singularity is approached, the arm velocities required for movement in the problematic direction approach infinity. As a practical matter, this means that the speed of movement of the tool 16 which the manipulator is capable of producing is greatly reduced as the arm approaches a singularity. While a redundant seven-axis machine has the inherent advantages over singularities, singularities still present a problem. For example, it is possible to encounter double singularities which eliminate two degrees of freedom, effectively reducing a seven-axis machine to a five-axis machine.

In addition to the problem of lost mobility presented by a singularity, the linear transformation equations used to determine the joint angle values have no mathematical solution at the point of singularity. Consequently, a digital computer called upon to calculate joint angle values using conventional numerical techniques typically will be unable to do so when it encounters a singularity.

As previously mentioned, the difficulty of joint angle calculation created by the presence of redundancy can be handled by introducing into the solution method an extra mathematical criteria for resolving the joint angle values. This extra criteria is called an objective function. An objective function is an equation which states a relationship between variables in a manner which permits the function to be solved for a maximum or minimum value, as specified, subject to certain constraints. In the preferred embodiment of the present invention, a convex objective function is used to deal with redundancy and cope with singularities which is based upon a "minimum spring energy" concept of manipulator operation. Under this approach, the extra mathematical criteria introduced into the algorithm used for calculating joint angle values is the requirement that each arm joint behave as though it were held to a fixed home position by a rotational spring coaxially mounted on the arm joint. This behavior can be achieved by specifying as the additional mathematical criteria that the joint angles be determined so as to minimize the value of a specific convex objective function subject to constraints derived from the transformation equations. That is, the constraint equations are linear approximations of the transformation equations.

Selection of a minimum spring energy convex objective function reference to a fixed origin has been found to provide numerous advantages. Most notably, use of such an objective function with the manipulator described above results in a conservative system which is repeatable to a designated point with a high degree of accuracy. Also, such an objective function causes the manipulator to distribute joint work and joint velocities among all arm joints in a reasonable manner. In addition, it tends to prevent the joints of the manipulator from reaching the limits of their permissible range of motion.

While the minimum spring energy convex objective function has been found to provide excellent operating characteristics of a jointed multi-axis manipulator arm, including improved repeatability, that discovery alone does not handle the problem of singularities. Sets of singularities still may be encountered in the operation of the manipulator and will result in an equation set with no mathematical solution. However, the preferred embodiment of the present invention incorporates a redundant joint algorithm which, together with the spring energy objective function, smoothly and effectively handles and tends to avoid sets of singularities.

The redundant joint algorithm is executed in the RCJC 274. It provides a method for calculating the joint angles required of the arm joints in order to achieve the next Intermediate Point calculated by the Trajectory Control 272.

As explained above, the equation set used to compute the joint angles consists of a convex objective function which is to be optimized subject to constraint equations. The objective function selected is based upon a physical analogy to spring energy but the redundant joint algorithm of the present invention will work effectively with any objective function expressed as a positive semi-definite quadratic form. More specifically, the equation set of the preferred embodiment uses a positive semi-definite quadratic form with a diagonal coefficient matrix as the objective function. The constraint equations of the equation set consist of linear approximations of the transformation equations--in other words, the Jacobian operator of the transformation equations. Thus, the mathematical problem to which the solution algorithm addresses itself may be stated using matrix notation as: ##EQU6## Where: n is the number of joints, normally seven or greater;

k.sub.i is the spring stiffness for the i.sup.th joint;

.theta..sub.i is the known current position of the i.sup.th joint;

.DELTA..theta..sub.i is the unknown change in angle for the i.sup.th joint from its current position;

.theta..sub.io is the known spring origin (i.e. the angle corresponding to zero hypothethical spring force) for the i.sup.th joint;

{.DELTA..theta.} is the vector of the unknown change in angle for each joint from its current position;

{.DELTA.X} is the vector of the change in cartesian coordinates of the tool position between the current position and the next position;

[J.degree.] is the Jacobian matrix evaluated at the current position. The Jacobian is an m.times.n matrix where m is the number of constraints (normally six or less) and n is the number of arm joints (normally greater than m). The Jacobian is defined by the following expression: ##EQU7## Where f is the vector of functions mapping joint variables into cartesian coordinates.

Given this statement of the problem, the first step in the solution algorithm involves a change of variables which results in the spring stiffnesses k.sub.i becoming unity and in the offsets of the joint angles from their spring origins (i.e. ({.DELTA..theta.}+{.theta.}-{.theta..sub.0 })) becoming zero. To accomplish this change of variables we let ##EQU8## That is, [K] is a diagonal matrix whose entries are one over the square roots of the spring stiffness k.sub.i. Then, we may let

The problem then may be restated as: ##EQU9## The problem is further simplified to finding a seven dimensional vector [D] that minimizes ##EQU10##

It can be proved that this objective function is minimized by choosing {D}=[F].sup.t {G}, where [F]=[E][A], [F][F].sup.t =[I] and {G}=[E]{B}; i.e., [F] is an orthonormal basis for the row space of [A].

Multiplying eq. (9) by matrix [E],

substituting [F]=[E][A] and {G}=[E]{B} we get

The proof of the proposition stated above goes as follows. Let [F.sup..perp. ] be an (n-m) by n matrix whose columns are orthogonal to [F] i.e.: [F].multidot.[F.sup..perp. ].sup.t =0). And let

where {H} is an (n-m) vector. To show that {D}.sup.* is a general solution to equation (10), substitute into {D}.sup.* into (10):

now using the properties for [F], namely

Therefore, {D}.sup.* is a general solution to the transformed constraint equation (10) where {H} is an arbitrary vector.

Given that, the problem is reduced to finding a value of {H} which minimizes ##EQU11## or equivalently 1/2D.sup.t D.

Substituting {D}.sup.* into the objective function,

expanding terms and using the properties for [F] and [F.sup..perp. ] previously defined, equation (13) reduces to ##EQU12## the minimum occurs when {H}=0.

Therefore, the objective function ##EQU13## is a minimum when {H}=0.

The matrix [E] does not need to be determined explicitly. Rather, if elementary row operations are performed on eq. (9) such that [A] is transformed into [F], then [B] is transformed into [G] without determining [E]. Furthermore, the matrix multiplication

can be performed concurrently with the matrix transformation of [A] to [F].

As stated above, the redundant joint algorithm executed by the RCJC 274 requires the application of a series of transformation to the matrix equation [A]{D}={B} so that it becomes [F]{D}={G} where [F] has orthonormal rows (i.e. [F][F].sup.t =[I]). This is done by subtracting from rows 2 through 6 of matrix [A] multiples of the first row so that the first row becomes orthogonal to the remaining rows. Next, multiples of row 2 are subtracted from rows 3 through 6 so that the second row now becomes orthogonal to the remaining row. This process is completed with each successive row of the matrix.

The factors used to compute the multiples of each row in the above process are calculated as follows. By definition, the factor .alpha. in the formula below is such that when a given row, a, is multiplied by the factor .alpha. and subtracted from a given row, b, rows a and b will be orthogonal. The orthogonality condition means that the dot product of the two vectors is zero. That is,

When the successive operations described above are performed on the matrix [A] and the vector {B}, the resulting matrix equation [F]{D}={G} provides the basis for computation of the required joint angles. Furthermore, this process provides a means for recognizing and treating a singularity. If the vector magnitude of one of the rows in matrix [A] becomes nearly zero during the execution of the steps described above, that indicates that a singularity has been encountered and a degree of freedom has been lost. When that condition arises, the nearly zero row of the [A] matrix being converted into [F] eliminated from the matrix and a minimum solution to the objective function is calculated with one less constraint equation.

After {D} is determined, a change of variables is performed using equation (5) providing the joint angles {.DELTA..theta.}. This solution method is iterative, converging on the final solution. To reach the final solution, therefore, the values of .DELTA..theta..sub.i determined at the end of the first iteration may be added to the values of .theta..sub.i and the algorithm may be repeated if greater accuracy is needed until .DELTA..theta..sub.i approaches zero. Portions of the listing of the RCJC program in the Appendix deviates somewhat from the purely mathematical description above.

Those skilled in the art will recognize that the redundant joint algorithm described above provides the ability to recognize the occurrence of singularities with relative computational ease and speed. Those skilled in the art will realize that the redundant joint algorithm will yield that benefit with any objective function which is expressed as a positive semi-definite quadratic form. Once the near singularity is detected as provided in the algorithm the problematic constraint equation is stricken from the constraint matrices. The effect of this is to reduce the number of known values in the equation set by one and thereby create a hypothetical added redundancy. Using ordinary mathematical solution techniques this would cause an imbalance between the number of equations and the number of unknowns which would be impossible to solve even with a manipulator consisting of six or less axes. However, the minimum spring energy convex objective function described above permits the modified set of equations (with the induced hypothetical redundancy) to be solved by application of the redundant joint algorithm.

In other words, those skilled in the art will recognize that this combination of objective function and solution technique also is quite effective for dealing with singularities arising in non-redundant manipulators. For example, a six-axis machine operating with all location values and tool orientation values specified is not operating in redundancy and therefore does not need the extra solution criteria provided by an objective function. Such a machine, nonetheless, will encounter singularities which are likely to make solution of the transformation equations impossible. With the redundant joint algorithm described above, the presence of a singularity will induce a hypothetical redundancy into the equation set. The hypothetical redundancy will be handled by the spring energy objective function or any other objective function expressed as a sum of squares using the redundant joint algorithm. Thus, the redundant joint algorithm of the present invention not only provides numerous improved operating characteristics and performance capabilities but also enables intelligent treatment of singularities in machines of any number of axes and does so with relative computational ease, speed and efficiency.

Returning to FIG. 10, the Servo Control Interface 276 receives from the RCJC 274 the joint angle values for each arm joint required to attain the next Intermediate Point. The Servo Control Interface 276 executes a series of calculations to determine whether the arm joints are physically capable of moving to the Intermediate Point within the 33 milli-second cycle time on which the first-tier of the controller 15 runs. The Servo Control Interface 276 accomplishes this calculation by comparing the next required Intermediate Point defined as joint angle values to the previously commanded joint angles. If one or more of the arm joints is incapable of achieving the next Intermediate Point joint angle value called for by the RCJC 274, the Servo Control Interface 6 suspends the operation of the 33 milli-second first-tier clock, preventing it from initiating the next round of computations in the Move Control 270, Trajectory Control 272 and RCJC 274. The Servo Control Interface 6 suspends operation of the clock a sufficient number of cycles to permit the problematic joint to attain the joint value position called for by the most recent command from the RCJC 274. At the same time, the Servo Control Interface 276 modifies the joint angle instructions to the remaining arm joints to ensure that they reach the next Intermediate Point simultaneously with the problematic arm joint. This safeguarding procedure is referred to as "scaling". With or without scaling, the Servo Control Interface 276 calculates the next resolver positions for each of the arm joints required to satisfy the Trajectory Control 272 and RCJC 74 and indicates the time allocated to reach the next resolver position. This information is supplied for all arm joints.

The next resolver position for each arm joint and the time allocated to attain that position are supplied to the Servo Control 278 from the Servo Control Interface 276. In contrast to the control system elements identified above, the Servo Control 278 of the preferred embodiment operates on a much faster clock cycle--in the range of 3 milli-seconds. For the controller 15 to operate properly, the slower clock must run at a speed which is an integer multiple of the faster clock. The Servo Control 278 receives constant input from each arm joint resolver indicating the present true position of that resolver and arm joint. By linear interpolation, the Servo Control 278 computes the resolver value for each arm joint that should be achieved at the end of each 3 milli-seconds cycle in order for each of the arm joints to achieve the resolver value and corresponding joint angles specified by the Servo Control Interface 276 by the end of the 33 milli-second time interval. The Servo Control 278 transmits to two Resolver Boards 296 & 298 a digital position error signal which is proportional to the difference between the most recent known joint positions and the joint positions required 3 milli-seconds later.

The present invention provides two Resolver Boards, 296 & 298 which are identical and perform identical functions. Two Resolver Boards 296 & 298 are provided to share the seven arm joints. The position of the Resolver Boards 296 & 298 in the flow of signals is illustrated in FIG. 10. The Resolver Boards 296 & 298 each receive a constant analog signal from the arm joint resolvers. The Resolver Boards 296 & 298 convert this to a digital signal indicative of the current resolver position and provide that information to the Servo Control 278 as described above.

Upon receiving the digital position error signals for each joint from the Servo Control 278 every 3 milli-seconds, the Resolver Boards 296 & 298 convert that signal to analog position error signals. This signal is transmitted to Analog Compensation Boards 282 and ultimately results in a drive signal to the joint motors. In addition, the Resolver Boards 296 & 298 generate a continuous analog signal indicative of the instantaneous angular velocity of each arm joint.

Turning now to FIG. 13, one of the Resolver Boards 296 & 298 is shown in greater detail. The Resolver Boards 296 & 298 are identical and each include the functions shown by the block diagrams in FIG. 13. The address decode 302 monitors address signals sent on the address bus 304 portion of a multibus 300, described above, and responds to a predetermined address signal which indicates that the Resolver Board and a particular component of the Resolver Board are to be accessed. The address decode 302 occurs in two parts. The first part is a base address which indicates that signals are to be received by the Resolver Board. Second, the address decode 302 selects the individual component contained on the Resolver Board which is designated by the address signal. Once an individual component of the Resolver Board is selected by the address decode 302, data transmission occurs via the databus buffers 306.

The digital position error signals generated by the Servo Control 278 mentioned above are transmitted sequentially to the digital to analog chips 308 ("D/A chips") contained on the Resolver Board. The D/A chips 308 are conventional commercially available items which convert a digital signal to a corresponding analog signal. One such signal is generated for each arm joint of the manipulator. Each analog signal is transmitted to the Analog Servocontrol Circuits 282 shown in FIGS. 10 and 15.

As previously explained, each arm joint includes a resolver. The resolver is a conventional commercially available device which provides two analog output signals indicative of the precise angular position of its arm joint. These signals are transmitted to the resolver to digital chips 310 ("R/D chips") provided on the Resolver Board. Each arm joint has a resolver and a corresponding R/D chip 310. The R/D chip 310 of the arm joint provides a continuous digital count derived from the analog resolver signals indicative of the resolver position and, thus, the arm joint angle. This digital count is transmitted via the databus buffers 306 and the multibus 300 to the Servo Control 278 and used, as described above, by the Servo Control 278 program to generate a digital position error signal. Each of the R/D chips 310 also provide an instantaneous analog velocity signal indicative of the instantaneous angular velocity of each arm joint which is transmitted to the Analog Compensation Boards 282.

Each R/D chip 310 has a specific tracking rate. The tracking rate is the rate at which the R/D chip 310 can process the analog signal received from the resolver and provide a corresponding digital value. When an arm joint moves at a speed which results in resolver signal variations which exceed the tracking rate, the R/D chip 310 generates a DC signal the magnitude of which is indicative of the lag between the actual resolver position and the instantaneous digital value produced by the R/D chip 310. This DC signal is transmitted for each arm joint to the error check 312. The error check 312 consists of a comparator circuit which compares the DC signal to a predetermined value. If the DC signal exceeds the predetermined value, the error check 312 latches a flip-flop on the status register 314. In the next cycle of digital signal transmission between the Resolver Board and the Servo Control 278, the Servo Control 278 receives a signal from the status register 314 via the databus buffers 306 informing it of the count lag and the Servo Control 278 responds accordingly.

Each of the R/D chips 310 also include a "busy line". When the busy line signal is on, it indicates that the R/D chip 310 is updating its digital count. The busy line and the count latches 320 (one is provided for each joint) can be used to prevent a read error resulting from a change in one or more of the digital values during a read. That is, the count latches 320 are set to update every time the busy line signal turns off from their associated R/D chip 310. In this manner, the processor reads the resolver counts directly from the count latches rather than from the R/D chip 310.

In certain applications, the manipulator of the present invention may be used for high accuracy metrology and the tool 16 carried on the end of the manipulator may be a touch probe. The count latches 320 may be used in such cases to capture the digital resolver values generated by the R/D chips 310 that existed at the moment the probe made contact with its target. In FIG. 13, a probe signal 316 is provided when the probe makes contact with its target. That signal is transmitted to a position latch 318. When the position latch 318 is triggered, a signal from the position latch 318 is transmitted to the count latches 320 and a signal is transmitted to the status register 314 to record the fact that a probe signal 316 was received. On the next cycle of the controller 15, the Servo Control 278 receives a signal from the status register 314 indicating that the probe has made contact. The Servo Control 278 program treats the digital signals received from the count latches associated with each joint in that cycle as an identification of the arm joint positions at the moment of probe contact. The position latch 318 then is unlatched and the count latches 320 proceed to update their digital values to match the R/D chips 310.

Refering again to FIG. 10 the Analog Servocontrol Circuits 282 receive from the Resolver Boards 296 and 298 an analog position error signal generated by the D/A chips 308 and an instantaneous velocity signal for each arm joint. The Analog Servocontrol Circuits 282 amplify the position error signal and, using the instantaneous velocity signal and the torque feedback signal explained below, compensate that signal to ensure stability--thereby providing a final drive signal to each arm joint motor.

The electromechanical system of a manipulator such as that of the present invention typically is a second order or higher order feedback control system, as those terms are used in the vernacular of control engineering. As such, the electromechanical system is susceptible to operational instability if the physical characteristics of the system and the drive signals provided to the joint motors are not properly controlled.

In many servomechanisms similar to the present invention, instability problems are minimized because the mechanical drive structure is very stiff and thus its resonant frequency is very high; beyond a point of serious concern. The forward path of the control loop of such a system may be compensated, (e.g. by a low frequency lag compensation network) thereby making the system behave as a first order system within the bandwidth of interest. This approach has the disadvantage of slowing the response of the servomechanism because it reduces the bandwidth of the system.

Such conventional servocontrol techniques are inappropriate with the manipulator described above. The mechanical drive system of the manipulator of the present invention is relatively compliant due to the flexibility of the harmonic drive. Consequently, the resonant frequency of the mechanical drive system is relatively low. Furthermore, the harmonic drive creates vibration and control problems that would be far less severe in a structure which used other power transmission means. That is, in addition to the typical excitation sources of drive signal frequency content and various mechanical roughnesses in the system, the harmonic drive causes sinusoidal excitation as a result of an inherent transmission deviation at the rate of n cycles per motor revolution where n is the number of lobes on the wave generators 100 or 260.

Because of these features, a typical servocontrol and compensation scheme consisting of a position feedback loop and a velocity feedback loop would be ineffective and sluggish if applied to the structure of the present invention. Conventional control strategies would not allow the resonance to be controlled and, in fact, may amplify the problem of the harmonic drive sinusoidal excitation.

To overcome these and related problems, the servocontroller of the present invention utilizes a torque feedback loop in addition to velocity and position feedback loops and in addition to a current loop included as part of a current amplifier. While all three feedback loops are used together in normal operation, the torque loop could be used alone for applications of the manipulator which call for application of a specified force to an object. Furthermore, the torque loop feedback control concept of the present invention could be advantageously employed in nearly any type of actuator driven mechanical apparatus; it is not limited to the revolute manipulator described above. By using the torque loop as the innermost loop, the motor and harmonic drive configuration of each arm joint operates as a torque producing plant rather than a motion producing plant. The torque feedback loop not only remedies the sinusoidal excitation to the structure which otherwise would be induced by the harmonic drive, it also improves the response of the servomechanism because it permits the position and velocity feedback loops to operate in a frequency range far beyond what would otherwise be possible. The use of the torque loop also significantly reduces the effects of friction and compliance in the servomechanism, enabling the present invention to operate with much greater accuracy and repeatability. Furthermore, the torque loop feedback improves the operating characteristics of the manipulator even when mounted on a soft foundation.

FIG. 14 is a block diagram of the present invention including the servocontrol elements. Each block of the block diagram contains the transfer function of the associated mechanical, electrical or electromechanical element of the entire system expressed as Laplace transforms. The symbols which correspond directly to physical properties of the portion of the system whose operation is being controlled are defined below.

L.sub.a --Armature Inductance

R.sub.a --Armature Resistance

K.sub.t --Motor Torque Constant

J.sub.m --Motor and Wave Generator Inertia

B.sub.m --Motor and Wave Generator Viscous Friction

K.sub.v --Voltage Constant

N--Drive Ratio

K.sub.d --Drive Spring Constant

J.sub.l --Joint Inertia

B.sub.l --Joint Viscous Friction

Those skilled in the art will recognize that the same or equivalent compensation systems could be represented by block diagrams different from that shown in the figure.

For simplicity, the portion of the block diagram contained within phantom box 322 may be viewed as the primary physical system of a single arm joint controlled by the servocontroller of the present invention. Blocks 324 and 326 represent the armature inductance and resistance and the torque constant of the drive motor, respectively. Block 328 represents the inertia and viscous friction of the drive motor and the harmonic drive wave generator. Block 330 represents the gear ratio of the harmonic drive and the time integration of velocity to position. Block 332 represents the spring constant associated with the harmonic drive and the other power transmission components leading up to the point where the strain gauges are applied.

Blocks 334, 336, and 338 and the associated feedback lines represent dynamic characteristics of a typical current amplifier/motor combination. Block 340 represents the axis torque reflected through the gear ratio to the motor.

As shown in the figure, the servocontrol of the present invention uses forward path compensation. Compensation also could be accomplished by placing compensation networks in the feedback loops to achieve the same functions.

As previously stated, the servocontrol of the present invention uses four feedback loops--a current loop 342, a torque loop 344, a velocity loop 346, and a position loop 348. The torque loop 344 represents the torque feedback. Block 350 represents the gain of the feedback transducer which is proportional to the torque in the drive at the point where the harmonic drive connects to the arm joint. Block 352 represents the torque loop compensation network.

Block 354 represents the inertia and friction of the manipulator arm segment. Velocity loop 346 represents the arm segment velocity feedback. Block 356 represents the gain of the velocity feedback transducer which is proportional to the velocity of the arm segment. Block 358 represents the velocity loop compensation network.

Finally, Block 360 represents the time integration of velocity to position. Block 362 is the gain of the position feedback transducer and 364 represents a gain in the forward path which determines the speed of response of the position loop. The arrow 366 represents the commanded position signal. The portion of the block diagram contained within the phantom box 368 represents functions which are performed in the digital portion of the controller 15.

Those skilled in the art will recognize that numerous alternative circuits and control schemes could be developed to accomplish the torque, velocity, and position and current feedback control of the present invention using conventional compensation design techniques such as those explained in Ogata, Modern Control Engineering, Prentice-Hall Inc. 1970, the relevant portions of which are incorporated herein by reference. If desired, for example, the values of the block diagram terms associated with the electromechanical system of each joint can be determined experimentally and used to mathematically derive an appropriate compensation network. Alternatively, the frequency response and phase characteristics of the portion of the system to be controlled can be experimentally determined and graphically represented by Bode plots, which would form the basis for design of appropriate forward path or feedback path compensation networks. Other design methods also could be employed.

The preferred embodiment of the servocontrol circuitry of the present invention is schematically illustrated in FIG. 15. For descriptive purposes, the schematic diagram may be logically divided into several portions outlined by the phantom boxes.

As indicated in the block diagram of FIG. 14 and shown in FIG. 15, the circuitry receives a position error signal from the D/A chips, a velocity signal from the R/D chips and a torque signal from the strain gauges. The position error signal is supplied to a differential amplifier amplifier shown in phantom box 370. Likewise, the velocity signal and torque signal are supplied to the differential amplifier and instrumentation amplifier shown in phantom boxes 372 and 374, respectively. Each of the differential amplifiers 370 and 372 includes an operational amplifier, indicated at 376 and 378, and an array of one or more resistors and capacitors. Instrumentation amplifier 374 is designed to provide greater gain than amplifiers 370 and 372 because the signals received from the strain gauges are quite small.

The signal provided by the differential amplifiers 370 and 372 is supplied to the velocity loop compensation network shown in phantom box 384 and adjusted by balance 386. Balance 386 permits the circuitry to be adjusted to provide zero output when there is zero position error.

The velocity loop compensation network 384 consists of a lag network of a particular type known as a lag-lead-lag network. A low frequency lag, consisting of an operational amplifier, resistor and capacitor, is provided to increase the signal gain at low frequencies, thus improving the static stiffness of the manipulator. This is followed by a medium frequency lead network, consisting of a resistor and capacitor, to improve stability and reduce overshoot of the manipulator. Finally, a high frequency lag network, consisting of an operational amplifier, resistor and capacitor, is provided as a filter. A velocity loop compensation network consisting of only a lag network or only a lag-lead network would also be functional but would not be as effective as the lag-lead-lag network illustrated.

The amplified torque signal from differential amplifier 374 is supplied to the lag network consisting of an operational amplifier, resistors and capacitors shown in phantom box 388 and is adjusted by balance 390. This network serves as a high frequency filter. The resulting signal is supplied to the torque loop compensation network shown in phantom box 394.

The torque loop compensation network 394 consists of a lead network of a particular type known as a lead-lag, preceded by a differential amplifier. A low frequency lead, consisting of a resistor and capacitor receives the signal from a differential amplifier. This is followed by a lag network consisting of an operational amplifier, resistor and capacitor. The resulting signal is supplied to a conventional current amplifier which produces the final drive signal to the joint motor. The purpose and effect of the torque loop compensation network is to alter the frequency response and phase characteristics of the open loop forward path to avoid a condition of instability and to maximize the operational bandwidth of this inner-most feedback loop. By maximizing the operational bandwidth of the torque loop, the bandwidth of the velocity and position loops are permitted to be higher than would otherwise be possible, thus improving the speed of response of the manipulator.

The values of the network components are determined experimentally or by mathematical or computer modeling. The values are unique for each structure being controlled. Therefore, the values are not indicated here.

The foregoing description of a preferred embodiment and alternative embodiments of the invention have been presented for purposes of illustration and description. The description is not intended to be exhaustive or to limit the invention to the precise form disclosed. Obviously, many modifications and variations are possible in light of the above teaching. The embodiment presented above was chosen and described in order to best explain the principles of the invention and its practical application to thereby enable others skilled in the art to best utilize the invention in various embodiments and with various modifications as are suited to the particular use contemplated. ##SPC1## 10/999