RoboSTD enables two-arm robot control from one-arm data
RoboSTD: Zero-Shot Single-to-Dual Transfer via Sagittal Mirroring for Bimanual Learning
Robotics
Summary
Robots that use both arms can do many tasks better, but it’s expensive to teach them by showing how to do these tasks with two arms. The authors found a way to use data recorded from one arm to train two-arm robots without extra examples. They copy the actions from one arm to the other side by flipping movements and create rules with the help of a large language model to coordinate the two arms. Their method works well in simulation and in real robots, helping a robot pick up more cups using two arms than when it uses only one.
What this means in practice
- •For robotics system developers: Train two-arm robot controllers using existing one-arm demonstration data without collecting costly new two-arm examples.
- •For warehouse automation integrators: Improve robotic cup or object collection tasks by enabling bimanual policies trained from single-arm data for faster deployment.$Commercial implications: Enables selling bimanual manipulation solutions with reduced training costs to logistics providers.
Authors
Hongyu Wu, Chong Liu, Shaoxuan Xie, Shaqi Luo, Hanyu Feng, Shihan Wu, Guocai Yao
Abstract
Bimanual robot manipulation plays an important role in real-world robotic applications, but the high cost of collecting bimanual demonstrations limits the scalable development of bimanual policy learning. Meanwhile, existing single-arm manipulation data are difficult to directly transfer to bimanual systems due to the Single-to-Dual gap in action spaces, coordination requirements, and arm-side spatial preference. To address this issue, we propose RoboSTD, a zero-shot Single-to-Dual data construction method for bimanual robot learning. RoboSTD first applies sagittal-plane mirroring to observations, proprioception, and actions to construct executable contralateral pseudo-demonstrations. It then uses an LLM to generate structured spatio-temporal coordination constraints that guide the reconstruction of pseudo-bimanual supervision across workspace regions and task stages. We evaluate RoboSTD on the AgileX ALOHA platform in both RoboTwin simulation and real-world settings. Experiments demonstrate the effectiveness of RoboSTD for bimanual policy learning. Without requiring additional bimanual demonstrations, RoboSTD increases the average number of collected cups to 2.11x that of the single-arm baseline in the cup collection task. Project page: https://huahungy.github.io/RoboSTD/