| --- |
| license: apache-2.0 |
| library_name: lerobot |
| pipeline_tag: robotics |
| datasets: |
| - RobotControlStack/duobench |
| base_model: |
| - lerobot/xvla-base |
| tags: |
| - duobench |
| - bimanual |
| - manipulation |
| - simulation |
| - xvla |
| --- |
| |
| # DuoBench X-VLA (sim) |
|
|
| X-VLA policy for the bimanual FR3 Duo setup of [DuoBench](https://arxiv.org/abs/2606.11901), trained on simulation data only for all 11 DuoBench sim tasks (one multi-task policy). |
| Fine-tuned from [`lerobot/xvla-base`](https://huggingface.co/lerobot/xvla-base) with [LeRobot](https://github.com/huggingface/lerobot) on the [DuoBench dataset](https://huggingface.co/datasets/RobotControlStack/duobench). |
|
|
| Project page: https://duobench.github.io/ |
| DuoBench Code: https://github.com/RobotControlStack/duobench |
|
|
| ## Training |
| - Data: 50 demonstrations per task from the `<task>/sim` folders of [RobotControlStack/duobench](https://huggingface.co/datasets/RobotControlStack/duobench) (all tasks merged) |
| - Inputs: `head`, `left_wrist` and `right_wrist` cameras (224x224) and the 16-D state (2 x 7 joints + 2 grippers); output: 16-D joint and gripper action |
| - 40,000 steps, batch size 48, learning rate 1e-4, action chunk 30, seed 1000 |
| - LeRobot commit [`c8ce413d`](https://github.com/huggingface/lerobot/tree/c8ce413d738da15a2eed2d0832315779ea28cbf9); the full configuration is in `train_config.json` |
|
|
| ## Results |
| Success rate (%) in DuoBench simulation over 300 rollouts per task (seeds 0-299) with a Wilson 95% interval. |
|
|
| | Task | Success | 95% CI | |
| |---|---|---| |
| | ball_maze | 2.0 | [0.9, 4.3] | |
| | bin_sort | 1.0 | [0.3, 2.9] | |
| | block_balance | 0.0 | [0.0, 1.3] | |
| | carry_pot | 12.3 | [9.1, 16.5] | |
| | hinge_chest | 12.0 | [8.8, 16.2] | |
| | join_blocks | 0.0 | [0.0, 1.3] | |
| | pour_marbles | 0.0 | [0.0, 1.3] | |
| | spring_door | 13.7 | [10.2, 18.0] | |
| | transfer_cube | 9.3 | [6.5, 13.2] | |
| | transfer_gate | 26.3 | [21.7, 31.6] | |
| | transfer_reorient | 0.0 | [0.0, 1.3] | |
| | **mean** | **7.0** | | |
| |
| ## Usage |
| ```python |
| # LeRobot at commit c8ce413d |
| from lerobot.policies.xvla.modeling_xvla import XVLAPolicy |
|
|
| policy = XVLAPolicy.from_pretrained("RobotControlStack/duobench-xvla-sim") |
| ``` |
| See the [DuoBench code](https://github.com/RobotControlStack/duobench) for the simulation environments and evaluation. |
| |
| ## Limitations |
| - Trained on simulation data only; not meant for the real robot. |
| - Loading needs a LeRobot version compatible with the commit above. |
| |
| ## Citation |
| If you find this model or DuoBench useful for your work, please consider citing it: |
| ``` |
| @misc{duobench, |
| title={{DuoBench}: A Reproducible Benchmark for Bimanual Manipulation in Simulation and the Real World}, |
| author={Tobias J{\"u}lg and Seongjin Bien and Simon Hilber and Yannik Blei and Pierre Krack and Maximilian Li and Sven Parusel and Rudolf Lioutikov and Florian Walter and Wolfram Burgard}, |
| year={2026}, |
| url={https://arxiv.org/abs/2606.11901} |
| } |
| ``` |
| |
