---
tags:
- robotics
- reinforcement-learning
- vision-language-action
- openpi
- bridge-data
---

# OpenPI 0.5 Bridge RL

This repository contains an OpenPI 0.5 policy checkpoint after reinforcement learning on Bridge-style manipulation tasks.

## Contents

- `model.safetensors`: policy model weights.
- `optimizer.pt`: optimizer state included for checkpoint compatibility.
- `metadata.pt`: training metadata with local filesystem paths, internal run identifiers, and wall-clock timestamps removed.

## Data

The associated reinforcement-learning data is available in `Fleurrr/Bridge-RL-Data`.

## Notes

- The checkpoint is provided in the original OpenPI training format.
- Loading and inference require a compatible OpenPI 0.5 codebase and model configuration.
- Internal infrastructure details and personal identifiers are intentionally omitted.