A General Technique to Combine Off-Policy Reinforcement Learning Algorithms with Satellite Attitude Control
Reinforcement learning method has great potential in constructing next generation of intelligent attitude control for satellite.However,designing reward function when using reinforcement learning algorithm to achieve specific task is a hard problem,which limits reinforcement learning algorithm used in satellite attitude control.For avoiding complicated reward engineering,we present a technique which allows the off-policy reinforcement learning algorithm can be easily migrated to construct satellite attitude control method.A satellite simulation environment is constructed.In this environment,we train an attitude control agent with the technique and validate the techniques effectiveness.
Attitude control Off-policy Reinforcement learning Reward function
Jian Zhang Fengge Wu Junsuo Zhao Fanjiang Xu
Institute of Software Chinese Academy of Sciences,Beijing,China
国际会议
江苏镇江
英文
709-719
2019-09-20（万方平台首次上网日期，不代表论文的发表时间）