This repository contains the datasets used in the paper From Prior to Pro: Efficient Skill Mastery via Distribution Contractive RL Finetuning.
Project Website | GitHub Repository
Distribution Contractive Reinforcement Learning (DICE-RL) is a framework that uses reinforcement learning (RL) to refine pretrained generative robot policies. This repository… See the full description on the dataset page:
https://huggingface.co/datasets/wintermelontree/robomimic-pretrain-data.
