Skill Policy

Skill policy research focuses on enabling robots to efficiently learn and execute complex tasks by decomposing them into reusable, parameterized skills. Current research emphasizes learning these skills from diverse data sources, including offline datasets and unstructured "play" data, often leveraging diffusion models and generative approaches to create robust and adaptable skill representations. This work aims to improve sample efficiency and generalization capabilities in reinforcement learning for robotics, leading to more adaptable and robust autonomous systems for real-world applications.