Robust Imitation Learning from Noisy Demonstrations
This addresses the problem of noisy demonstrations in imitation learning for robotics or AI systems, offering a robust solution without extra labels or strict noise assumptions, though it appears incremental as it builds on existing theoretical frameworks.
The paper tackles robust imitation learning from noisy demonstrations by proposing a method based on optimizing a classification risk with symmetric loss, combining pseudo-labeling and co-training, which outperforms state-of-the-art methods on continuous-control benchmarks.
Robust 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.