Experimental Assessment of Deep Reinforcement Learning for Robot Obstacle Avoidance: A LPV Control Perspective

This work presents the experimental assessment of a hybrid control scheme based on Deep Reinforcement Learning (DRL) for obstacle avoidance in robot manipulators. More precisely, relying an equivalent Linear Parameter Varying (LPV) state-space representation system, two operative modes, one both joint positions and velocities, only velocity inputs, are activated depending measurement distance between obstacle. Therefore, when is close to robot, switching mechanism introduced enable DRL algorithm instead basic motion planner, thus giving rise self-configuring architecture cope with objects randomly moving workspace. The tests collision strategy carried out physical EPSON VT6 manipulator satisfactory results.