AnyBimanual: Transferring Single-arm Policy for General Bimanual Manipulation
Keywords: Multi-task Bimanual Manipulation, Skill Learning
Abstract: Performing language-conditioned bimanual manipulation tasks is of great importance for many applications ranging from household service to industrial assembly. However, teleoperating dual-arm demonstrations is expensive due to the high-dimensional action space, which poses challenges for conventional methods to handle general bimanual manipulation tasks. In contrast, single-arm policy has recently demonstrated impressive generalizability across a wide range of tasks because of scaled model parameters and training data, which can provide sharable manipulation knowledge for dual-arm systems. To this end, we propose a plug-and-play method named AnyBimanual, which transfers pretrained single-arm policy to multi-task bimanual manipulation policy with limited bimanual demonstrations. Specifically, we first introduce a skill manager to dynamically schedule the discovered skill primitives from pretrained single-arm policy for bimanual manipulation tasks, which combines skill primitives with embodiment-specific compensation. To mitigate the observation discrepancy between single-arm and dual-arm systems, we present a voxel editor to generate spatial soft masks for visual embedding of the workspace, which aims to align visual input of single-arm policy model for each arm with those during pretraining stage. Extensive results on 13 simulated and real-world tasks indicate the superiority of AnyBimanual with an improvement of 12.67\% on average success rate compared with previous state-of-the-art methods.
Supplementary Material: zip
Primary Area: applications to robotics, autonomy, planning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 6293
Loading