Potential Driven Reinforcement Learning for Hard Exploration Tasks
文献类型：会议论文
| 作者 | Zhao EM(赵恩民)1,2 ; Deng SH(邓诗弘)1,2; Zang YF(臧一凡)1,2; Kang YX(康永欣)1,2; Li K(李凯)1,2; Xing JL(兴军亮)1,2
|
| 出版日期 | 2020-04 |
| 会议日期 | 2020-4 |
| 会议地点 | 线上 |
| DOI | 无 |
| 英文摘要 | Experience replay plays a crucial role in Reinforce
ment Learning (RL), enabling the agent to remem
ber and reuse experience from the past. Most pre
vious methods sample experience transitions us
ing simple heuristics like uniformly sampling or
prioritizing those good ones. Since humans can
learn from both good and bad experiences, more
sophisticated experience replay algorithms need to
be developed. Inspired by the potential energy in
physics, this work introduces the artifificial potential
fifield into experience replay and develops Potential
ized Experience Replay (PotER) as a new and ef
fective sampling algorithm for RL in hard explo
ration tasks with sparse rewards. PotER defifines a
potential energy function for each state in experi
ence replay and helps the agent to learn from both
good and bad experiences using intrinsic state su
pervision. PotER can be combined with different
RL algorithms as well as the self-imitation learning
algorithm. Experimental analyses and comparisons
on multiple challenging hard exploration environ
ments have verifified its effectiveness and effificiency |
| 语种 | 英语 |
| URL标识 | 查看原文 |
| 源URL | [http://ir.ia.ac.cn/handle/173211/52252]  |
| 专题 | 融合创新中心_决策指挥与体系智能 |
| 通讯作者 | Xing JL(兴军亮) |
| 作者单位 | 1.School of Artificial Intelligence, University of Chinese Academy of Sciences 2.Institute of Automation, Chinese Academy of Sciences |
| 推荐引用方式 GB/T 7714 | Zhao EM,Deng SH,Zang YF,et al. Potential Driven Reinforcement Learning for Hard Exploration Tasks[C]. 见:. 线上. 2020-4. |
入库方式： OAI收割
来源：自动化研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。