Jehuda Ish-Shalom

SPARTA: multiple signal processors for high-performance robotic control

The authors describe the SPARTA (signal processor architecture for real-time applications) system, which is a hybrid computer development system including: (1) a program development environment on an IBM VM/CMS mainframe computer: (2) user interface and runtime support on an IBM PC; and (3) real-time computation and input/output using multiple IBM Hermes signal processors situated on the IBM PC bus. The program development environment includes the PLH high-level language for generating efficient real-time Hermes code. The runtime supports symbolic debugging, dynamic loading and linking, and synchronous switching of control algorithms during real-time program execution. The real-time environment includes a distributed operating system which supports foreground and background task execution and real-time data collection and display. The task switching overhead is 0.1 mu s, nonpreemptive, and 1.0-3.7 mu s when the context is changed by an interrupt (i.e. preemptive). >

The CS Language Concept: A New Approach to Robot Motion Design

Current robot control techniques are primarily concerned with position control and, recently, with force control, where a motion planner is used primarily to compute set point changes. The actual control system is fixed and internal and cannot be modified easily through software. Current robot languages are also limited in their ability to specify changes to the control structure. For compliant motion, such as as sembly, more flexibility is needed in the control system, that is, an ability to tailor the controller to the task. This paper develops a high-level control system (CS) language that allows one to specify compliant control tasks as vector equa tions and inequalities that relate sensed and controlled vari ables such as f · v = 0 and f X v = 0 . We show how such requirements can be reformulated as an objective function of an optimization process subject to constraints arising from the dynamic equations involved. Specifically for the two examples above, we show how such requirements can be re formulated as quadratic objective functions and solved using standard linear optimization theory for linear dynamic sys tems. Six examples are presented to illustrate the various ideas.

Experimental results of using a linear step motor as a programmable spring

It is demonstrated experimentally that a variable-reluctance linear step motor can be controlled as a programmable spring with a spring constant of dynamic range 2000 to 1 for robot applications. This large dynamic range is mainly attributed to the direct-drive use of this variable-reluctance linear step motor. Using direct drive provides a high ratio of output force to friction because of the lack of commutation brushes and a reduction gear. The large dynamic range of achievable spring constants also demonstrates that a very simple electronic commutation-logic feedback can simplify a statically and dynamically complex variable-reluctance step motor to the point that it can be algebraically linearized. For a large operating range in force, velocity, and frequency, this linearized model can be approximated by a second-order ordinary differential equation; i.e. the force on a mass given by Newtons second law with the addition of a small friction force.<<ETX>>

The CS language concept: A new approach to robot motion design

Current robot control techniques are primarily concerned with position control and recently also with force control, where a motion planner is used primarily to compute set point changes. The actual control system is fixed and internal, and cannot be modified easily through software. Current robot languages are also limited in their ability to specify changes to the control structure. For compliant motion, such as assembly, one needs more flexibility in the control system; i.e. an ability to tailor the controller to the task. In this paper, we develop a high level CS language (Control System) which allows one to specify compliant control tasks as vector equations and inequalities which relate sensed and controlled variables, e.g. f? ? v? = 0 and f? × v? = 0?. We show how such requirements can be reformulated as an objective function of an optimization process subject to constraints arising from the dynamic equations involved. Specifically for the two examples above, we show how such requirements can be reformulated as quadratic objective functions and solved using standard linear optimization theory for linear dynamic systems. Five examples are presented to illustrate the various ideas.

Signal processor architecture for high-performance real-time applications

A description is given of the software environment for SPARTA (signal processor architecture for real-time applications). The SPARTA hardware consists of a hybrid system with three different types of computers. The program development environment includes the PLH high-level language (on an IBM VM/CMS mainframe) for generating efficient real-time signal processor code. The runtime (user) interface (on an IBM PC) supports symbolic debugging, dynamic loading and linking, and synchronous switching of control algorithms during real-time program execution. It has been interfaced to the object-oriented AML/2 language interpreter. The real-time environment (on multiple IBM Hermes signal processors) has tasks with multiple entries and exits, linked by a method which is somewhat similar to threaded code, but which has lower overhead (0.1 mu s, nonpreemptive task switch). The real-time kernel also supports multiple threads of tasks categorized as critical or abortable, reflecting their importance to the real-time system. These threads are scheduled by a priority-based, preemptive scheduler, which has 1.3-5.1 mu s total overhead. A SPARTA system with one Hermes signal processor, executing a dozen real-time tasks, has achieved a sample rate of 5 kHz for a fully-coupled PID control, with nonlinear compensation, for the Eaglet II Cartesian robot (five motors achieving 2- mu m linear resolution and 0.02 degrees angular resolution).<<ETX>>

Task independent relationship of system to actuator performance via scaling

The authors present a scaling with which it is possible to predict the relative savings with respect to the time required for planar motions when using different actuators. This calculation can be done independently of the sequence of motions to be performed; in particular, when the sequence of motions is not yet known. This is commonly the case when the motion system is being designed. The use of this produce independent analysis requires that the scaling can only be done with respect to one free parameter representing the actuator. The analysis has the advantage that it simplifies comparison between actuators, since the relative performance is represented by a single scalar. It is shown that with the proper scaling the optimal sequence will remain unaffected. The improvement in the systems performance is quantifiable and independent of the distribution of the distances which must be traversed.<<ETX>>