Robust Imitation Learning from Noisy Demonstrations
Robust Imitation Learning from Noisy Demonstrations
复制标题
DOI：
--
复制
发表时间：
2020-10
期刊：
影响因子：
--
通讯作者：
Voot Tangkaratt;Nontawat Charoenphakdee;Masashi Sugiyama
中科院分区：
文献类型：
--
作者：
Voot Tangkaratt;Nontawat Charoenphakdee;Masashi Sugiyama
从嘈杂的演示中学习是模仿学习中一个实际但极具挑战性的问题。在本文中，我们首先从理论上证明，可以通过优化具有对称损失的分类风险来实现鲁棒的模仿学习。基于这一理论发现，我们提出了一种新的模仿学习方法，通过有效地将伪标签与协同训练相结合来优化分类风险。与现有方法不同，我们的方法不需要额外的标签或关于噪声分布的严格假设。连续控制基准的实验结果表明，与最先进的方法相比，我们的方法更加稳健。
Learning from noisy demonstrations is a practical but highly challenging problem in imitation learning. In this paper, we first theoretically show that robust imitation learning can be achieved by optimizing a classification risk with a symmetric loss. Based on this theoretical finding, we then propose a new imitation learning method that optimizes the classification risk by effectively combining pseudo-labeling with co-training. Unlike existing methods, our method does not require additional labels or strict assumptions about noise distributions. Experimental results on continuous-control benchmarks show that our method is more robust compared to state-of-the-art methods.