Implementation of a multiprocessor system for real-time inverse dynamics computation

The implementation of a 32-bit multiple microprocessor-based system for real-time inverse dynamics computation is presented. The recursive Newton-Euler algorithm for serial-link robotic manipulators is decomposed into subtasks, which are then arranged for parallel computation using scheduling algorithms. The hardware system uses currently available 32-bit microprocessors and provides easy expandability for accommodating multiple microprocessors. The implementational and experimental results for computing the inverse dynamics algorithms are presented. Using the six-axis Stanford manipulator as an example, the actual computing time for executing the complete inverse dynamics is within 1.758 ms. The execution time can be further reduced by customizing the scheduling algorithms, with a resultant time of 1.472 ms. The results indicate that real-time computation of sophisticated robotic control algorithms can be realized using currently available high-performance 32-bit microprocessors