The article introduces SLIM, a multi-critic learning approach for skill discovery, particularly focusing on robotic manipulation. It addresses the limitations of mutual information maximization in producing useful manipulation behaviors. The approach uses multiple critics in an actor-critic framework to combine multiple reward functions effectively, leading to significant improvements in latent-variable skill discovery. It also shows the applicability of this skill discovery approach to acquire safe and efficient motor primitives in a hierarchical reinforcement learning fashion. The method surpasses the state-of-the-art approaches for skill discovery by a large margin.
Publication date: 2 Feb 2024
Project Page: N/A
Paper: https://arxiv.org/pdf/2402.00823
