README.md

metadata

license: apache-2.0
library_name: lerobot
pipeline_tag: robotics
datasets:
  - RobotControlStack/duobench
tags:
  - duobench
  - bimanual
  - manipulation
  - simulation
  - act

DuoBench ACT (sim)

ACT policy for the bimanual FR3 Duo setup of DuoBench, trained on simulation data only for the DuoBench pour_marbles sim task (single-task policy). Trained from scratch with LeRobot on the DuoBench dataset.

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
• 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
• 150,000 steps, batch size 64, learning rate 1e-5, action chunk 100, seed 1000
• LeRobot commit c8ce413d; 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
pour_marbles	0.3	[0.1, 1.9]

Usage

# LeRobot at commit c8ce413d
from lerobot.policies.act.modeling_act import ACTPolicy

policy = ACTPolicy.from_pretrained("RobotControlStack/duobench-act-sim-pour_marbles")

See the DuoBench code 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}
}