Active Domain Randomization
This addresses the issue of poor generalization in domain transfer for robotics and simulation tasks, offering an incremental improvement over standard domain randomization.
The paper tackled the problem of domain randomization leading to suboptimal, high-variance policies by proposing Active Domain Randomization, which learns a parameter sampling strategy to focus on informative environment variations, resulting in more robust and consistent policies across simulated and real-robot tasks.
Domain randomization is a popular technique for improving domain transfer, often used in a zero-shot setting when the target domain is unknown or cannot easily be used for training. In this work, we empirically examine the effects of domain randomization on agent generalization. Our experiments show that domain randomization may lead to suboptimal, high-variance policies, which we attribute to the uniform sampling of environment parameters. We propose Active Domain Randomization, a novel algorithm that learns a parameter sampling strategy. Our method looks for the most informative environment variations within the given randomization ranges by leveraging the discrepancies of policy rollouts in randomized and reference environment instances. We find that training more frequently on these instances leads to better overall agent generalization. Our experiments across various physics-based simulated and real-robot tasks show that this enhancement leads to more robust, consistent policies.