RoboSTD: Zero-Shot Single-to-Dual Transfer via Sagittal Mirroring for Bimanual Learning

Hongyu Wu; Chong Liu; Shaoxuan Xie; Shaqi Luo; Hanyu Feng; Shihan Wu; Guocai Yao

cs.ROOct 27, 2026

RoboSTD: Zero-Shot Single-to-Dual Transfer via Sagittal Mirroring for Bimanual Learning

Authors: Hongyu Wu, Chong Liu, Shaoxuan Xie, Shaqi Luo, Hanyu Feng, Shihan Wu, Guocai Yao

Organizations: Beijing Academy of Artificial Intelligence · Beijing University of Posts and Telecommunications

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/

Figures & tables

Fig. 1: Overview and motivation of RoboSTD. RoboSTD transforms spatially biased single-arm demonstrations into pseudo-bimanual supervision through two stages: sagittal-plane mirroring constructs executable contralateral skills, while LLM-guided spatio-temporal reconstruction coordinates skills through arm assignment, precedence, and conflict constraints, broadening workspace coverage and mitigating arm-side spatial preference.

Fig. 2: Arm-side spatial preference analysis in robot data and policies. (a) End-effector workspace distribution estimated from Open X-Embodiment trajectories to characterize workspace bias. (b) Left- and right-arm performance comparison of pretrained

\pi_{0}

and non-pretrained ACT across 10 RoboTwin 2.0 tasks, examining whether data-level spatial imbalance is reflected in downstream policy performance.

Method	Source Supervision	No Additional Bimanual	Automated Data Generation	Sagittal Mirroring	Single-to- Dual	Spatio-Temporal Supervision
DexMimicGen	Bimanual demos	$\times$	$\checkmark$	$\times$	$\times$	$\circ$
D-CODA	Bimanual demos	$\times$	$\checkmark$	$\times$	$\times$	$\times$
ROPA	Bimanual demos	$\times$	$\checkmark$	$\times$	$\times$	$\times$
AnyBimanual	Pretrained unimanual policy	$\times$	$\times$	$\times$	$\checkmark$	$\circ$
RoboSTD	Single-arm demos	$\checkmark$	$\checkmark$	$\checkmark$	$\checkmark$	$\checkmark$

TABLE I: Comparison of Data Reuse and Single-to-Dual Transfer Methods for Bimanual Robot Learning

Fig. 3: Overview of RoboSTD. (a) Inputs include single-arm demonstrations, robot geometry, and task context. (b) Sagittal-plane mirroring jointly transforms observations, proprioception, and actions to generate contralateral pseudo-demonstrations. (c) An LLM infers arm assignments, precedence, and conflict constraints. (d) These constraints guide the composition of original and mirrored skills into pseudo-bimanual supervision.

Fig. 4: RoboSTD evaluation on 10 RoboTwin tasks. (a) Dual-arm AgileX ALOHA platform and 10 manipulation tasks covering rotation, pressing, positioning, and placement. (b) Task-wise success rates of $\pi_{0}$ under Single-Arm, RoboSTD, and Oracle supervision settings.

Fig. 5: Real-world evaluation of RoboSTD across coordination levels. (a) Low-coordination tasks covering rigid, deformable, and precision manipulation. (b) High-coordination tasks comparing Single-Arm, RoboSTD w/o LLM, RoboSTD, and Oracle to evaluate structured spatio-temporal coordination using task-specific metrics.

Fig. 6: RoboTwin evaluation of arm-side spatial preference and its mitigation with RoboSTD. (a–b) End-effector trajectory density and arm-side success rates of the Single-Arm-finetuned baseline. (c–d) Corresponding results of the RoboSTD-based policy on the three selected RoboTwin tasks.

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

Fig. 7: Examples of mirror-consistent joint-sign assignments under sagittal reflection. For representative symmetrically mounted bimanual systems, the dashed line indicates the sagittal plane $\mathcal{P}_{s}$ . Blue joints denote sign-preserving mappings ( $\lambda_{i}=+1$ , $q_{i}^{L}=q_{i}^{R}$ ), whereas red joints denote sign-inverting mappings ( $\lambda_{i}=-1$ , $q_{i}^{L}=-q_{i}^{R}$ ). The same sign-assignment principle applies consistently across the illustrated robot morphologies. The assignment is determined by the reflected joint-axis relation and the robot mounting configuration, rather than by the joint name itself.

Bimanual Robotic Manipulation

Bimanual robotic manipulation coordinates two robot arms to handle objects, tools, and tasks that require collaboration.

See all 189 papers

MonoDuo: Using One Robot Arm to Learn Bimanual Policies
May 28, 2026Sandeep Bajamahal, Lawrence Yunliang Chen, Toru Lin +3Bimanual Robotic Manipulation Cross-Embodiment Robot Learning
University of California, Berkeley
Bimanual Robot Manipulation via Multi-Agent In-Context Learning
Apr 22, 2026Alessio Palma, Indro Spinelli, Vignesh Prasad +4ICL for Robotics Language Model-Based Control
Sapienza University of Rome, Italy · TU Darmstadt, Germany · Hessian.AI, Germany
One-to-Two Acting: A Novel Framework for Single-arm Agent Action Expansion to Dual Arms
Jun 18, 2026Youbin Yao, Nieqin Cao, Mingyan Li +3Language-Conditioned Robot Manipulation Bimanual Robotic Manipulation
1Chongqing University, Chongqing, China · 2Xi’an Jiaotong-Liverpool University, Suzhou, China · 3Lumos Roboticss

May 28, 2026J/K move · Enter open · S save

MonoDuo: Using One Robot Arm to Learn Bimanual Policies

Sandeep Bajamahal, Lawrence Yunliang Chen, Toru Lin, Zehan Ma+2

University of California, Berkeley

Bimanual Robotic Manipulation Cross-Embodiment Robot Learning Robot Imitation Learning

Open paper AskPDF arXiv

Bimanual coordination is essential for many real-world manipulation tasks, yet learning bimanual robot policies is limited by the scarcity of bimanual robots and datasets. Single-arm robots, however, are widely available in research labs. Can we leverage them to train bimanual robot policies? We present MonoDuo, a framework for learning bimanual manipulation policies using single-arm robot demonstrations paired with human collaboration. MonoDuo collects data by teleoperating a single-arm robot to perform one side of a bimanual task while a human performs the other, then swapping roles to cover both sides. RGB-D observations from a wrist-mounted and fixed camera are augmented into synthetic demonstrations for target bimanual robots using state-of-the-art hand pose estimation, image and point cloud segmentation, and inpainting. These synthetic demonstrations, grounded in real robot kinematics, are used to train bimanual policies. We evaluate MonoDuo on five tasks: box lifting, backpack packing, cloth folding, jacket zipping, and plate handover. Compared to approaches relying solely on human bimanual videos, MonoDuo enables zero-shot deployment on unseen bimanual robot configurations, achieving success rates up to 70%. With only 25 target robot demonstrations, few-shot finetuning further boosts success rates by 65-70% over training from scratch, demonstrating MonoDuo's effectiveness in efficiently transferring knowledge from single-arm robot data to bimanual robot policies.

Abstract

Bimanual Robotic Manipulation

Explore similar work

MonoDuo: Using One Robot Arm to Learn Bimanual Policies

Bimanual Robot Manipulation via Multi-Agent In-Context Learning

One-to-Two Acting: A Novel Framework for Single-arm Agent Action Expansion to Dual Arms