Towards Vision-Based Deep Reinforcement Learning for Robotic Motion Control

• DOI: 10.48550/arxiv.1511.03791
• Publication Date: 2015-01-01
• Authors: Fangyi Zhang; Jürgen Leitner; Michael Milford; Ben Upcroft; Peter Corke
• Keywords: Visual control; Manipulator (device)

This paper introduces a machine learning based system for controlling robotic manipulator with visual perception only. The capability to autonomously learn robot controllers solely from raw-pixel images and without any prior knowledge of configuration is shown the first time. We build upon success recent deep reinforcement develop target reaching three-joint using external observation. A Deep Q Network (DQN) was demonstrated perform after training in simulation. Transferring network real hardware observation naive approach failed, but experiments show that works when replacing camera synthetic images.