The mapping of an adaptive robot control algorithm onto RM, a reconfigurable machine for highly parallel real-time applications

Abstract

The paper describes a configuration of the reconfigurable machine (RM) specifically optimized to handle inverse dynamics computation for robot control and to implement a neural network control strategy in a parallel platform. The difficulties of operating on a reconfigurable platform are overcome by defining a fixed processing element (PE) model with multiple processing components and by using the PE's local memory for storing micro-instructions to drive the parallel processors. Traditionally, reconfigurable platforms have been used to contain all the execution sequences and a local memory is usually provided to store the necessary data. VHDL description of the model has been used for simulation, synthesis and optimization for mapping to XILINX 4005 FPGA technology.<>