3D RL-DWA

3D RL-DWA: A Hybrid Reinforcement Learning and Dynamic Window Approach for Goal-Directed Local Navigation in Multi-DoF Robots

Announce Type: replace Abstract: In this paper, we present a novel hybrid approach that combines Reinforcement Learning (RL) with Dynamic Window Approach (DWA) for adaptive 3D local navigation of high-degree-of-freedom robotic systems. Our method leverages sparse point cloud data to dynamically adjust both the motion and the shape of a deformable microrobot, enabling the system to navigate toward a goal in complex, constrained environments while maximizing the occupied volume. We evaluate...