---
license: mit
tags:
- robotics
- robot-learning
- preference-data
pretty_name: RoboArena Dataset Snapshot — 2026-07-17
---

# RoboArena Dataset Snapshot — 2026-07-17

This dataset contains autonomous policy rollouts, task-success
scores, policy-preference annotations, and related metadata collected by the
[RoboArena benchmark](https://robo-arena.github.io/) from its inception through July 17, 2026.

## Snapshot statistics

- Evaluation sessions: 3,883
- Policy episodes: 10,783
- MP4 videos: 27,148
- NPZ proprioception/action files: 10,783
- Session metadata files: 3,883

## Data layout

```text
.
├── README.md
├── global_metadata.yaml
└── evaluation_sessions/
    └── <evaluation_session_uuid>/
        ├── metadata.yaml
        ├── <label>_<policy_name>/
        │   ├── ..._video_left.mp4       # when available
        │   ├── ..._video_right.mp4      # when available
        │   ├── ..._video_wrist.mp4
        │   └── ..._npz_file.npz
        └── ...
```

Each session's `metadata.yaml` contains its instruction, preference,
long-form feedback, timestamps, and per-policy outcome fields. The top-level
`global_metadata.yaml` contains aggregate counts and a policy index with
open-source and action-space annotations.

Policy names are preserved exactly inside metadata. For portable directory
names, characters outside letters, numbers, `.`, `_`, and `-` are replaced by
underscores. For example, `allenai/MolmoAct2-DROID` is stored under a directory
component named `allenai_MolmoAct2-DROID`.

For details about the benchmark, see **“RoboArena: Distributed Real-World
Evaluation of Generalist Robot Policies,”**
[arXiv:2506.18123](https://arxiv.org/abs/2506.18123).