Background
Rising environmental complexity delays these calculations, compromising robot safety and reliability during tasks. Given the growing prevalence of Neural Processing Units (NPUs) in embedded systems, this project explores methods to parallelize and optimize current motion planning algorithms, eliminating redundant computations for step determination.
Research Objectives
We will optimize motion planning algorithms for complex environments on parallel processors. Finally, the optimized algorithms will be deployed onto embedded vector processing units or NPUs, significantly reducing motion planning latency in complex environments through hardware-software co-design.
Methods
Collision-free motion planning, Deformable Object Manipulation
Innovation
Motion planning on CPU-NPU heterogeneous System
Expected Outcomes
Minimize the robotic motion planning latency on dynamic obstacles