Task-Relevant Adversarial Imitation Learning
This addresses a critical vulnerability in imitation learning for robotics, enabling better task performance without task rewards, though it is incremental as it builds on existing adversarial methods.
The paper tackled the problem of adversarial imitation learning where discriminator networks learn spurious associations, leading to poor task performance, and proposed TRAIL, which outperformed standard GAIL and other baselines in robotic manipulation tasks from pixels.
We show that a critical vulnerability in adversarial imitation is the tendency of discriminator networks to learn spurious associations between visual features and expert labels. When the discriminator focuses on task-irrelevant features, it does not provide an informative reward signal, leading to poor task performance. We analyze this problem in detail and propose a solution that outperforms standard Generative Adversarial Imitation Learning (GAIL). Our proposed method, Task-Relevant Adversarial Imitation Learning (TRAIL), uses constrained discriminator optimization to learn informative rewards. In comprehensive experiments, we show that TRAIL can solve challenging robotic manipulation tasks from pixels by imitating human operators without access to any task rewards, and clearly outperforms comparable baseline imitation agents, including those trained via behaviour cloning and conventional GAIL.