A 1200×1200 8-Edges/Vertex FPGA-Based Motion-Planning Accelerator for Dual-Arm-Robot Manipulation Systems

Abstract

An SoC-FPGA-based motion-planning accelerator operating on a graph with >10M edges is presented for the first time for dual-am-robot manipulation systems. The proposed modified A∗ algorithm with minimized memory access time, is further accelerated by extensive parallel computation and dynamic reconfigurations. The proposed accelerator has been verified by measurement result showing overall motion-planning time of 0.5 seconds, which is only 1/100 of the one by conventional algorithm on embedded CPU, while preserving accuracy.