RoboArena: Distributed Real-World Evaluation of Generalist Robot Policies
Paper
•
2506.18123
•
Published
The dataset is currently empty. Upload or create new data files. Then, you will be able to explore them in the Dataset Viewer.
This dataset contains autonomous policy rollouts, task success scores, policy preference annotations, and related metadata collected by the RoboArena benchmark from its inception through February 3, 2026.
DataDump_02-03-2026/
├── global_metadata.yaml
└── evaluation_sessions/
├── <eval_session_id_0>/
│ └── …
├── <eval_session_id_1>/
│ └── …
└── <eval_session_id_n>/
├── metadata.yaml
├── A_<policy_A_name>/
│ ├── left_shoulder.mp4
│ ├── right_shoulder.mp4
│ ├── wrist.mp4
│ └── proprio_and_actions.npz
├── B_<policy_B_name>/
│ └── …
└── C_<policy_C_name>/ # If more than two policies
└── …
For more details, see “RoboArena: Distributed Real-World Evaluation of Generalist Robot Policies,” arXiv:2506.18123.