A seven-degree-of-freedom (7-DOF) anthropomorphic arm requires the control of the relative orientation of the elbow to satisfy task, kinematic, and dynamic constraints. While prior work has dealt primarily with iterative repulsive schemes to avoid joint limit stops, a geometric algorithm for explicitly computing the set of feasible configurations of a zero-offset arm at an arbitrary position with respect to the current position is presented. The data facilitates trajectory planning that takes into account a wide range of factors, including the limit stops, rather than reacting to limit stops encountered during trajectory execution. The algorithm was implemented to run on the JIFFE robotics processor. JIFFE computes the feasible configuration set and, mostly for testing, uses the data for inverse kinematics in 10 mu s.>

A system design is described for robotics research, including rapid program development and concurrent and post execution analysis. The system design supports robotic systems controlling 24 or more joints at a 1-kHz servo rate, suitable for controlling ensembles of advanced arms or hands. The system performs trajectory generation, kinematics, dynamics, force control, and advanced servo algorithms at the servo rate. The JIFFE robotics processor provides the computational power, performing 20 MFlops on robot control algorithms written in C. A sophisticated semiautonomous interface card for eight joints, designed to minimize system latency, is also described. A multiprocessor PUMA controller was ported onto this architecture. The reduced code requirements and the performance, featuring a 56- mu s sensor/actuator latency, are described.>

The author considers the resources required to implement robot trajectory, kinematic, dynamic, and servo algorithms at rates high enough to ensure that the computation rate does not limit performance. He examines current computer architectures in general and those intended for robot control. He then describes a very-long-instruction-word (VLIW) processor designed for robot control. This processor, called JIFFE, was designed to run close to its peak performance even on the random scalar operations characteristic of robotics algorithms. JIFFE performs nearly 20 MFLOPS (million floating-point operations per second) on practical scalar robotics code, while maintaining high performance on poorly structured code and executing some matrix code at 40 MFLOPS. Because JIFFE was designed to be a good research tool, it can be easily programmed in C. A variety of compiler features help generate efficient code, even from voluminous machine-generated expressions.>

A robot ping-pong system which attains good performance in a complex nonlinear environment subject to tight temporal constraints is described. It is shown how the expert controller combines approximate estimates and feedback to arrive at a suitable task plan, emphasizing feasibility rather than strict optimality. The characteristics of the expert controller that cause it to work and their relationship to other tasks are discussed. Particular attention is given to the use of task redundancy, continual integration of new sensor data, continual improvement of the task plan, fast approximate methods to generate values, physics-based models to ensure accuracy, global exception handling, flexible internal architecture and robustness to failure.>

The application of sensing technologies to an environment that changes at rates comparable to the sensing and actuation rates of a robot system is considered. In such dynamic environments, the temporal characteristics of the sensor must be taken into account. Additionally, the processing system must be able to integrate the stream of sensor data into its task plan, compensating for changed sensor data and continuously optimizing the plan. The author examines these issues in the context of a functioning robot system that plays ping-pong.>

A trajectory generation and tracking system is described which is designed to maximize the manipulator's usable performance on a sensor-driven task. The system drives a PUMA 260 that is part of a functioning robot ping-pong player. The system's implementation and performance in practice are reported.>

A system for performing visual processing on gray-scale images in real time (60 Hz) has been constructed. The custom VLSI moment generator chip computes area, center of gravity, orientation, and size. An image preprocessor allows separate moments to be computed for separate regions. A standard set of buses allows new processing elements to be easily added. The system is expected to find application in real-time sensor-based electronic assembly and in automated inspection and registration tasks. A simple application to a real-time visually servoed robot task is summarized.