arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 1806.02454cs.RO
收藏

从人类纠正中学习时包括不确定性

Including Uncertainty when Learning from Human Corrections

Rice University(莱斯大学)

Dylan P. Losey, Marcia K. O'Malley

摘要:

人类很难有效地教机器人如何正确地执行任务。一个直观的解决方案是机器人从修正中迭代地学习人类的偏好,其中人类在每次迭代时改进机器人的当前行为。当从修正中学习时,我们认为,虽然机器人应该估计最有可能的人类偏好,但它也应该知道它不知道什么,并在做出决策时整合这种不确定性。我们通过引入卡尔曼滤波器从校正中学习来推进最先进的技术:这种方法获得了估计的人类偏好的不确定性。接下来,我们将演示如何利用估计不确定性进行主动学习和风险敏感部署。我们的研究结果表明,获得和利用不确定性可以更快地从人类纠正中学习。

英文摘要:

It is difficult for humans to efficiently teach robots how to correctly perform a task. One intuitive solution is for the robot to iteratively learn the human's preferences from corrections, where the human improves the robot's current behavior at each iteration. When learning from corrections, we argue that while the robot should estimate the most likely human preferences, it should also know what it does not know, and integrate this uncertainty as it makes decisions. We advance the state-of-the-art by introducing a Kalman filter for learning from corrections: this approach obtains the uncertainty of the estimated human preferences. Next, we demonstrate how the estimate uncertainty can be leveraged for active learning and risk-sensitive deployment. Our results indicate that obtaining and leveraging uncertainty leads to faster learning from human corrections.

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。

↑