Parallel Computation Systems For Robotics Algorithms And Arch PDF Download

As a result of the automation revolution, robots are assuming ever more complex and demanding tasks. Robot system control schemes for fast and precise robot motion require utilization of the entirety of the robot dynamic formulations and the ability to evaluate these formulations in real-time. The dynamic formulations, which take into account robot nonlinearities and dynamic coupling, are computationally intensive. They are difficult to implement in real-time at high sampling rates due to the time required to compute the dynamic formulations. Reducing the computation time for practical implementation can be achieved by developing a computing algorithm for the efficient evaluation of the dynamic formulations and by designing a dedicated computer architecture. The proposed solution is a pipelined multiprocessor computer architecture and fast parallel algorithms for real-time control. The multiprocessor system can be utilized to concurrently perform pipelined parallel computations, thereby substantially increasing controller processing speed and CPU utilization. Concurrent performance of pipelined parallel computations is based on consideration of the sequential dependencies of the dynamic formulations which are conducive to pipelining, and decomposition of the dynamic backward formulations for fast parallel computation. The decomposition of the backward formulations is based on computational simplification techniques. The performance of the proposed algorithms, called "PAFP," is evaluated through analytic error analysis and experimental simulations, including motion simulations. It is compared to other approaches to the problem proposed by, respectively, Bejczy and Binder. Study results show that the errors introduced by decomposition are relatively small and compare very favorably to those obtained by other methods of computation simplification. The proposed computer architecture and the algorithms may be implemented with multiple low-cost microprocessors. This will allow a practical implementation of a highly parallel structure to achieve real-time robot control with high sampling rates.