Coupled Distributional Random Expert Distillation for World Model Online Imitation Learning
This addresses instability issues in imitation learning for robotics and autonomous systems, representing an incremental improvement over adversarial methods.
The paper tackles instability in online imitation learning by proposing a reward model based on random network distillation for density estimation, achieving expert-level performance with improved stability across benchmarks like DMControl and Meta-World.
Imitation Learning (IL) has achieved remarkable success across various domains, including robotics, autonomous driving, and healthcare, by enabling agents to learn complex behaviors from expert demonstrations. However, existing IL methods often face instability challenges, particularly when relying on adversarial reward or value formulations in world model frameworks. In this work, we propose a novel approach to online imitation learning that addresses these limitations through a reward model based on random network distillation (RND) for density estimation. Our reward model is built on the joint estimation of expert and behavioral distributions within the latent space of the world model. We evaluate our method across diverse benchmarks, including DMControl, Meta-World, and ManiSkill2, showcasing its ability to deliver stable performance and achieve expert-level results in both locomotion and manipulation tasks. Our approach demonstrates improved stability over adversarial methods while maintaining expert-level performance.