---
license: mit
task_categories:
- robotics
---

# Combining Constrained Sampling and Reinforcement Learning for Manipulation

This repository contains data associated with the paper [Combining Constrained Sampling and Reinforcement Learning for Manipulation](https://huggingface.co/papers/2602.08557).

- Project page: https://www.user.tu-berlin.de/mtoussai/26-CSRL/
- Code: https://github.com/LIS-TU-Berlin/csrl

The paper studies goal-conditioned reinforcement learning for contact-rich manipulation, combining constrained sampling with projected interpolation and curriculum learning to improve exploration and policy learning.