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

# Bridge RL Data

This dataset contains Bridge-style manipulation prompts and initial-state images used for reinforcement-learning experiments.

## Structure

The repository includes three JSON Lines manifests and an `images/` directory:

- `dataset.jsonl`: 403 records.
- `dataset_40.jsonl`: 680 records.
- `dataset_only_spoon.jsonl`: 1,268 records.
- `images/`: 10,212 PNG images.

Each manifest record contains:

- `task_type`: task or scene identifier.
- `task_language`: natural-language instruction.
- `init_state_path`: relative path to the initial-state image.

## Privacy

The published manifests use relative paths. Local filesystem locations, account names, machine identifiers, and embedded image metadata are not included.

## Related Model

The associated policy checkpoint is available in `Fleurrr/OpenPI05-Bridge-RL`.