LEARNING GOAL-CONDITIONED VALUE FUNCTIONS WITH ONE-STEP PATH REWARDS RATHER THAN GOAL- REWARDS

最新推荐文章于 2024-05-20 19:39:47 发布

Adam婷

最新推荐文章于 2024-05-20 19:39:47 发布

阅读量889

点赞数

CC 4.0 BY-SA版权

分类专栏：论文研读 ICLR 深度强化学习强化学习

本文链接：https://blog.youkuaiyun.com/weixin_41697507/article/details/94278689

强化学习同时被 3 个专栏收录

26 篇文章 ¥19.90 ¥99.00

订阅专栏

超级会员免费看

论文研读

38 篇文章

订阅专栏

深度强化学习

19 篇文章

订阅专栏

本文提出了一种新的目标条件强化学习算法，该算法通过学习一步路径奖励而非目标奖励来去除目标奖励的依赖，提高了样本效率，尤其是在奖励计算方面。通过对Fetch-Push任务的分析，证明了目标奖励的冗余性，同时扩展了Floyd-Warshall强化学习算法以适应深度神经网络。实验结果显示，这种方法在性能上与HER相当，但减少了奖励计算的需求。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

学习目标条件的价值功能与一步走的路径奖励比目标奖励更多

ABSTRACT

Multi-goal reinforcement learning (MGRL) addresses tasks where the desired goal state can change for every trial. State-of-the-art algorithms model these problems such that the reward formulation depends on the goals, to associate them with high reward. This dependence introduces additional goal reward resampling steps in algorithms like Hindsight Experience Replay (HER) that reuse trials in which the agent fails to reach the goal by recomputing rewards as if reached states were psuedo-desired goals. We propose a reformulation of goal-conditioned value func-tions for MGRL that yields a similar algorithm, while removing the dependence of reward functions on the goal. Our formulation thus obviat