Servo-Controlled Mobility Device

Subsequent to World War II there have been extensive efforts to control automatically functions which previously required the continuous intervention or control of a human operator. Numerical Control Servo System, Forrester et al., U.S. Pat. No. 3,069,608, directed to digitally controlled apparatus, is an example of a sophisticated version of such techniques. However, techniques such as the Forrester digitally controlled milling machine resulted in so-called open-loop apparatus, that is, the apparatus traveled along a specified path, but interaction between the apparatus and its environment or objects along its path played no part in determining the course of the tool. Fundamentally similar industrial systems are now on the market in the form of American Machine Foundry's "Versatran" and Unimation Inc.'s "Unimate". These machines are all basically open-loop machines under the control of pre-programmed instructions with minimal, if any, feedback. They may, for example, employ some feedback to stop further motion when the head comes into position to make, for example, a desired weld.

Certain researchers began to realize that a more sophisticated system was necessary if every application was not to involve very specific, specialized predesigned programming. At Massachusetts Institute of Technology's Draper Laboratory extensive work was done upon systems involving force and torque feedbacks to "steer" a mechanical arm or hand to provide desired functions. For example the Master of Science thesis of Roland C. Groome Jr., (Massachusetts Institute of Technology, 1972, also available as MIT C. S. Draper Laboratory report T-575) based upon work done under Mr. James L. Nevins at the Draper Laboratory, describes in considerable detail a system involving force steering loops to control an Argonne National Laboratories Model E-2 manipulator, a device having six different degrees of freedom. The apparatus is capable of functions such as inserting a half-inch diameter peg into a hole with a clearance between the peg and the hole of only approximately five thousandths of an inch. It became apparent, however, that even apparatus such as Groome's involved considerable design that was specific to a given task.

Accordingly it is an object of the present invention to achieve motion control apparatus which is adapted to the performance of many different tasks through easily performable modifications to a single portion of the control apparatus.

A further object is to achieve motion control apparatus wherein changes in the environment or modifications to the sensing apparatus may be readily incorporated without modifying other parts of the system. A further object is to provide a motion control system which may be easily applied to mechanical arms or other controllable motion apparatus through easily accomplished system adjustments. A further object is to achieve motion control apparatus in which each part of the system may be dealt with in its own natural or most convenient coordinates, including the coordinates of the environment, so as to facilitate achievement of the above objectives.

These and other objects of the invention are achieved in a servo system which includes a servo loop incorporating the controllable element, resolved motion rate control apparatus or other coordinate transformation apparatus as appropriate, and an accommodation branch, all of whose transfer characteristics may be readily adjusted to constrain the controllable element to accomplish a desired task.

FIG. 1 is a schematic representation of a system embodiment incorporating a mechanical arm.

FIG. 2 is a mathematical model of a system embodiment incorporating a mechanical arm.

FIG. 3 is a schematic representation of coordinates for a gripper for use in the system of FIGS. 1 or 2.

FIG. 4 is a schematic representation of force components for a force sensor for use in the systems of FIGS. 1 or 2.

FIG. 5 is a schematic representation of an embodiment of the systems of FIGS. 1 and 2.

FIG. 1 shows schematically the physical elements of an example system constructed according to the above objectives, carrying out an example task. It consists of a controllable motion device 20, namely an Argonne arm, a servo 22 to drive that arm, coordinate conversion apparatus 24, force and torque sensing apparatus 26 at or near the end of the arm, (such sensing apparatus could be located in the environment, or sensory apparatus could be located both on the controlled element and in the environment), a peg 28 grasped by the arm, and a hole 30 in a block located in the environment 32. Higher level task commands 34 indicate to the computer 36 that the peg 28 is to be put in the hole 30. Commands x.sub.D (21) represent velocity commands expressed in the coordinate system of the arm's gripper 38, and are designed to drive the peg to the mouth of the hole 30. The coordinate conversion apparatus 24 converts these commands into .theta..sub.D (23), the desired commands expressed in the coordinates of the degrees of freedom of the arm. When the tip of the peg hits the mouth of the hole, forces and torques are exerted on the peg. These forces and torques are sensed by the sensing apparatus and expressed in the coordinates of the gripper. The sensing apparatus signals are sent to the accommodation branch 40 which, according to task related logic, generates velocity modification commands X.sub.DM (25) in gripper coordinates. The net velocity commands are then converted, as before, into motion commands in the controllable degrees of freedom of the arm. In some systems constructed according to these principles, the computer 36 may perform the accommodation, force sensor processing, coordinate conversion and/or servo functions, but this is not necessary.

FIG. 2 shows a mathematical model of a servo controlled motion device in elastic contact with a generalized environment. This model is useful for designing such systems and understanding their behavior. That portion of the system within the dotted box 2 represents the controllable motion device with its control and transfer function terms representing the controllable element, the environment and the sensing device. That portion within the dotted box 4 represents the accommodation branch which includes transfer functions which will constrain the system's operation to achieve the desired task, along with functions which mathematically represent the environment and the sensing device characteristics. Box 6 represents a resolved motion rate conroller which serves to produce signals appropriate to the desired task motion when its output is applied to a controllable motion device of an articulated nature, or other complex motion device. If the motion device is particularly simple, the coordinate conversions accomplished by box 6 may not be necessary.

The various portions of the servo system of FIG. 2 will now be discussed in greater detail. Box 10 contains a typical transfer function which mathematically represents the controllable element - for example, a mechanical arm such as the Argonne arm - interacting elastically with its environment. (Other types of interaction impedance could be utilized, but this interaction is used as in example.) I.sub.a represents the inertia matrix of the controllable element while K.sub.E represents the stiffness matrix of the force sensors and the environment. If, for example, an Argonne arm is used as a controllable element, matrix I.sub.a may be calculated as set forth in Kahn, M. E., "The Near Minimum-Time Control of Open Loop Articulated Kinematic Chains," Ph.D Thesis, Stanford University, 1970. (Available from Xerox University Microfilms, Ann Arbor, Michigan, order No. 70-18425.) I.sub.a is expressed in the coordinates of the controllable degrees of freedom of the arm. Matrix K.sub.E is calculated as set forth in Appendix 1, and is expressed in this example in the coordinates of the gripper to which the sensor is attached. The remaining terms in box 10 are the Laplace operator s, the matrix J representing the resolved motion rate control function and superscript T representing matrix transpose. Matrix J may be calculated as set forth in Whitney, D. E., "The Mathematics of Coordinated Control of Prosthetic Arms and Manipulators", ASME Journal of Dynamic Systems, Measurement and Control, December, 1972, pp 303-309.

Box 6 performs the resolved motion rate control function and acts to convert a vector of motion commands x.sub.D (1) or x.sub.DM (15) (expressed in task-related coordinates) into commands to the controllable degrees of freedom of the motion device (for example an Argonne arm) so that the desired motions of the controllable motion device will occur. (In FIG. 2 these motions are expressed in the coordinates of the controllable degrees of freedom of the motion device (.theta.)(5) for mathematical convenience.) FIG. 3 shows schematically a typical gripper for an arm such as the Argonne arm with a typical set of coordinates attached to it.

Block 12 contains a typical servo for an arm such as the Argonne arm. This helps determine the response of the controllable element servo loop. Matrices K.sub.T and K.sub.TD are chosen to provide the type of characteristics which the servo loop will exhibit. For example, one may adjust or tune the servo to control the sensitivity and rate at which a particular task is accomplished. The coefficients are calculated as set forth in Nevins, J. L., Whitney, D. E., and Simunovic, S. N., "System Architecture for Assembly Machines", C. S. Draper Laboratory Inc. Report No. R-764, pp. 39-53. In some arm mechanizations, the actuators may be sufficiently responsive so that the servo loop closed around box 2 and the servo apparatus in box 12 are not needed, or are substantially simplified. In other cases, more complex servo apparatus may be needed in box 12, and should be designed according to the art of servo control design.

Considering now the accommodation branch in dotted box 4, the task logic which determines the way the controllable element will be constrained to perform a task is determined by the transfer function in box 14. If the controllable motion device is an Argonne arm, F(13) is a vector of forces and torques applied to the gripper by the environment. Matrix K.sub.F is an admittance matrix which receives input F (13) and converts it into velocity modifications vector x.sub.DM (15), both expressed in gripper coordinates. Force and torque vector F arises from deformations .DELTA.x (11) which occur in the environment and in the force sensor as a result of contact between the controllable element and the environment. .DELTA.x is also expressed in gripper coordinates. For mathematical convenience, this deformation is shown in FIG. 2 as .DELTA..theta. (9), its equivalent in the coordinates of the controllable degrees of freedom of the controllable element. Matrix J in box 18 is the same as the matrix which is used in box 6. Box 18 is part of the mathematical model only and has no counterpart in a physical system. The coordinates of the environment are also shown for mathematical convenience in the coordinates of the controllable element as vector .theta..sub.R (7). The sign conventions in this branch are such that, if K.sub.F is taken to be the identity matrix, then if a force vector F (13) is applied to the gripper, the vector x.sub.DM (15) will be exactly parallel to and proportional to vector F. Matrix K.sub.E (Box 16) is the same as the matrix K.sub.E which appears in box 10. It mathematically models the force-deformation characteristics of the sensor and the environment.

The calculation of matrix K.sub.F is considerably eased by the fact that its input and output are both in gripper coordinates. K.sub.F may be calculated as set forth for example in Nevins, J. L., and Whitney, D. E., "The Force Vector Assembler Concept," MIT C. S. Draper Lab. Report No. E-2754, the admittance matrix shown therein serving as K.sub.F for a system designed to put a peg in a hole as shown at page 13 of Nevins and Whitney. As set forth in this report, it is assumed that the gripper coordinates are velocities as follows: ##EQU1## and that the desired task is to put a peg in a hole starting from a point where the peg is in the mouth of the hole but is misaligned. The task logic is to use the forces and torques created by a command in the reach direction to create velocity modifications which will tend to correct the misalignment. Therefore, set ##EQU2## and let K.sub.F be the matrix

This matrix will cause motions other than reach to occur in accommodation to the cross forces and torques caused by misalignment. However, the 1 in the upper left corner of the matrix (in Report E-2754, the sign of this 1 is incorrect) will cause little modification to the reach motion unless forces sensed in the reach direction become extremely high, such as when the peg bottoms in the hole, at which point motion will stop. Except for the zeroes, the quantities in the matrices set forth in the specification are representative and may be adjusted along with command vector x.sub.D to modify the precise characteristics of system performance, including the overall servo dynamics, the force levels at which motion will commence or cease, and so on. The sign conventions may also be adjusted at will, so long as the accommodation branch and the remainder of the apparatus remain a stable dynamic system.

In some mechanizations of such systems, it will be convenient or appropriate to observe the motions of the arm via sensors on the arm or in the environment, and compose torque or force commands to the arm's actuators according to similar task related logic. This will be useful in such tasks as stretching springs or cables to a specified degree of deformation. In other mechanizations, it will be useful or convenient to sense the forces and torques applied to the arm by the environment and compose force or torque commands to the arm via an accommodation branch according to task related logic. This will be useful in such tasks as applying desired force or torque patterns. In each case if the driving components or arm have a response sufficiently precisely related to the command inputs it may be unnecessary to have the feedback connection 17 shown in FIG. 2, in which case servo 12 may be substantially simplified or altered.

FIG. 5 shows in block diagram form a detailed embodiment of the system of FIG. 1, which also corresponds to the mathematical model shown in FIG. 2. The computer subroutines indicated in FIG. 5 are shown in Appendix II.

Matrix K.sub.E represents the stiffness of the force sensing apparatus and the environment, and is defined mathematically by the relation

where C.sub.S represents the compliance of the sensor and C.sub.EN represents the compliance of the environment. We shall pursue here the case where the environment is, relative to the sensor, infinitely stiff, so that C.sub.EN = 0. Then K.sub.E = K.sub.S = C.sub.S.sup.-1.

The stiffness matrix K.sub.S of the force sensor shown schematically in FIG. 4 and discussed in Groome has, by design, the form

This is a diagonal matrix. The meaning of each entry is: ##EQU3## F.sub.x, F.sub.y, M.sub.x, M.sub.y, M.sub.z are defined in FIG. 4 and comprise the force vector applied to a point on the sensor when it contacts the environment, and is the force vector which the sensor is intended to measure. .delta..sub.x, .delta..sub.y and .delta..sub.z are linear displacements of this contact point relative to the base of the sensor, while .theta..sub.x, .theta..sub.y and .theta..sub.z are relative angular displacements. These displacments occur when the force vector is applied at the contact point and is counterbalanced by forces and moments which support the base of the sensor, the latter being held by the controllable element or arm. In FIG. 4, the contact point is a distance "a" from the base, as shown. The four bars of length "b" shown in this figure will deform elastically and counterbalance the applied force vector. Following the principles of structural analysis (see for example Crandall and Dahl, editors, "Introduction to the Mechanics of Solids", McGraw Hill, 1959, Chapter 8) one may calculate the entries in matrix K.sub.S, assuming that each bar of length "b" has an area moment I and a Young's modulus E, with these results: ##EQU4##

The formulae for K.sub..theta.x and K.sub..theta.z are obtained on the assumption that b is substantially less than a, the usual case for the device shown in FIG. 4. When any such device is built, it is advisable to determine the entries of K.sub.S experimentally as well as theoretically, to determine their values and the degree to which the intended diagonality of K.sub.S has been achieved. To do this, one must physically apply known loads and movements and, by means of suitable guages, measure the resulting displacements. ##SPC1## ##SPC2## ##SPC3## ##SPC4##