The Realization of Mobile Robot’s Dynamic Obstacle Avoidance with Deep Reinforcement Learning Based on Deterministic Strategy Gradient

DOI

10.2991/ncce-18.2018.200How to use a DOI?

Keywords

Deep Reinforcement Learning, Mobile Robot, Intelligent Decision-making, Visual Control, Local path planning.

Abstract

When the deep reinforcement learning algorithm based on visual perception is applied to the issue of robot’s dynamic obstacle avoidance, the perception ability of convolutional neural network is combined with the decision control ability of reinforcement learning, and the process from raw visual input to decision action output is realized. But the application scope of deep reinforcement learning algorithm based on Q value is still in low-dimensional and discrete action space. If the continuous action space is discretized, the problem of excessively huge motion space and extremely difficult convergence of network model will be caused. Besides, fine adjustment cannot be realized for the network model, and meanwhile, the division of motion space will also result in information loss. Hence, a deep reinforcement learning algorithm based on deterministic strategy gradient was proposed in this paper, and the strategy was parameterized via convolutional neural network through the integration of reinforcement learning algorithms based on strategy and value.

Copyright

© 2018, the Authors. Published by Atlantis Press.

Open Access

This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

TY  - CONF
AU  - Yunxiong Wu
PY  - 2018/05
DA  - 2018/05
TI  - The Realization of Mobile Robot’s Dynamic Obstacle Avoidance with Deep Reinforcement Learning Based on Deterministic Strategy Gradient
BT  - Proceedings of the 2018 International Conference on Network, Communication, Computer Engineering (NCCE 2018)
PB  - Atlantis Press
SP  - 1182
EP  - 1185
SN  - 1951-6851
UR  - https://doi.org/10.2991/ncce-18.2018.200
DO  - 10.2991/ncce-18.2018.200
ID  - Wu2018/05
ER  -