《Towards Vision-Based Deep Reinforcement Learning for Robotic Motion Control DQN ROS 1511.03791v2》.pdf

Towards Vision-Based Deep Reinforcement Learning for Robotic Motion Control Fangyi Zhang, Jürgen Leitner, Michael Milford, Ben Upcroft, Peter Corke ARC Centre of Excellence for Robotic Vision (ACRV) Queensland University of Technology (QUT) fangyi.zhang@.au 5 1 Abstract 0 2 This paper introduces a machine learning based v system for controlling a robotic manipulator o with visual perception only. The capability N to autonomously learn robot controllers solely from raw-pixel images and without any prior 3 1 knowledge of configuration is shown for the first time. We build upon the success of recent deep ] reinforcement learning and develop a system G for learning target reaching with a three-joint L robot manipulator using external visual obser- . s vation. A Deep Q Network (DQN) was demon- c strated to perform target reaching after train- Figure 1: Baxter’s arm being controlled by a trained [ ing in simulation. Transferring the network to deep Q Network (DQN). Synthetic images (on the right) 2 real hardware and real observation in a naive are fed into the DQN to overcome some of the real-world v approach failed, but experiments show that the issues encountered, i.e., the differences between training 1 network works when replacing camera images and testing settings. 9 7 with synthetic images. 3 0