arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2101.04178cs.ROcs.LG
收藏

机器人大动作空间的动作先验

Action Priors for Large Action Spaces in Robotics

Northeastern University(东北大学)

Ondrej Biza, Dian Wang, Robert Platt, Jan-Willem van de Meent, Lawson L. S. Wong

摘要:

在机器人技术中,如果没有显著的奖励塑造或课程学习,通常不可能使用纯无模型强化学习来学习有用的策略。因此,许多研究人员依靠专家示范来指导学习。然而,获得专家演示可能是昂贵的。本文提出了一种替代方法,其中以前解决的任务的解决方案被用来产生一个行动之前,可以促进探索在未来的任务。行动先验是行动的概率分布,它总结了解决先前任务的策略集。我们的研究结果表明,这种方法可以用来解决机器人操作的问题,否则是不可行的,没有专家的示范。源代码可从\url{https://github.com/ondrejba/action_priors}获得。

英文摘要:

In robotics, it is often not possible to learn useful policies using pure model-free reinforcement learning without significant reward shaping or curriculum learning. As a consequence, many researchers rely on expert demonstrations to guide learning. However, acquiring expert demonstrations can be expensive. This paper proposes an alternative approach where the solutions of previously solved tasks are used to produce an action prior that can facilitate exploration in future tasks. The action prior is a probability distribution over actions that summarizes the set of policies found solving previous tasks. Our results indicate that this approach can be used to solve robotic manipulation problems that would otherwise be infeasible without expert demonstrations. Source code is available at \url{https://github.com/ondrejba/action_priors}.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑