---
license: other
tags:
- robotics
- vla
- reinforcement-learning
- pi0.5
---

# SimpleVLA Pi0.5 GRPO Checkpoints

This repository contains released checkpoints for the SimpleVLA-style Pi0.5 GRPO experiments.

## Contents

- `pi05_droid_jointpos_pytorch_new/`: initial Pi0.5 DROID JointPos PyTorch checkpoint used for RL.
- `droid_grpo_best_stage03_global_step_5_3scene256_success48p44_20260602/`: locally selected best GRPO checkpoint.

The code, Docker setup, and download scripts live in the GitHub project repository.