Flow-based Domain Randomization for Learning and Sequencing Robotic Skills
This work addresses the challenge of improving robustness in robotic control policies for simulation-to-real transfer, representing an incremental advancement over manual or simpler automated domain randomization techniques.
The paper tackles the problem of automatically discovering sampling distributions for domain randomization in reinforcement learning by proposing a normalizing-flow-based neural sampling distribution optimized via entropy-regularized reward maximization. It demonstrates greater robustness than existing methods in six simulated and one real-world robotics domain, and explores its use for out-of-distribution detection in a manipulation planner.
Domain randomization in reinforcement learning is an established technique for increasing the robustness of control policies trained in simulation. By randomizing environment properties during training, the learned policy can become robust to uncertainties along the randomized dimensions. While the environment distribution is typically specified by hand, in this paper we investigate automatically discovering a sampling distribution via entropy-regularized reward maximization of a normalizing-flow-based neural sampling distribution. We show that this architecture is more flexible and provides greater robustness than existing approaches that learn simpler, parameterized sampling distributions, as demonstrated in six simulated and one real-world robotics domain. Lastly, we explore how these learned sampling distributions, combined with a privileged value function, can be used for out-of-distribution detection in an uncertainty-aware multi-step manipulation planner.