TY - GEN
T1 - Autonomous grading work using deep reinforcement learning based control
AU - Nakatani, Masayuki
AU - Sun, Zeyuan
AU - Uchimura, Yutaka
N1 - Funding Information:
This research is supported by the Space Exploration Innovation Hub Center of Japan Aerospace Exploration Agency, JAXA, whose activity is selected for the support program for starting up innovation hub sponsored by the Japan Science and Technology Agency (JST).
Publisher Copyright:
© 2018 IEEE.
PY - 2018/12/26
Y1 - 2018/12/26
N2 - The field of artificial intelligence (AI) has advanced significantly over the years. One of its achievements is the deep reinforcement learning algorithm using which AI can play some Atari 2600 games better than humans. In this paper, optimal route of construction machines such as bulldozers is modeled based on deep reinforcement learning. The aim of this study is to apply deep reinforcement learning to a grading machine to enable it to grade various surface types autonomously. A simple grading simulator is created to simulate the grading task. In addition, the overall scenario is made visible to the network by entering the simulation into the network so that human operators can construct suitable ground path from the surrounding sediment environment. The method is evaluated with the grading simulator, and the agent is shown to exhibit desirable control behavior and fulfill the goals of the simple grading simulation. Despite the environment being virtual, the simulation results demonstrate the feasibility of the proposed approach.
AB - The field of artificial intelligence (AI) has advanced significantly over the years. One of its achievements is the deep reinforcement learning algorithm using which AI can play some Atari 2600 games better than humans. In this paper, optimal route of construction machines such as bulldozers is modeled based on deep reinforcement learning. The aim of this study is to apply deep reinforcement learning to a grading machine to enable it to grade various surface types autonomously. A simple grading simulator is created to simulate the grading task. In addition, the overall scenario is made visible to the network by entering the simulation into the network so that human operators can construct suitable ground path from the surrounding sediment environment. The method is evaluated with the grading simulator, and the agent is shown to exhibit desirable control behavior and fulfill the goals of the simple grading simulation. Despite the environment being virtual, the simulation results demonstrate the feasibility of the proposed approach.
UR - http://www.scopus.com/inward/record.url?scp=85061556154&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85061556154&partnerID=8YFLogxK
U2 - 10.1109/IECON.2018.8591189
DO - 10.1109/IECON.2018.8591189
M3 - Conference contribution
AN - SCOPUS:85061556154
T3 - Proceedings: IECON 2018 - 44th Annual Conference of the IEEE Industrial Electronics Society
SP - 5068
EP - 5073
BT - Proceedings
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 44th Annual Conference of the IEEE Industrial Electronics Society, IECON 2018
Y2 - 20 October 2018 through 23 October 2018
ER -