MonoDuo: Using One Robot Arm to Learn Bimanual Policies

Sandeep Bajamahal, Lawrence Yunliang Chen, Toru Lin, Zehan Ma, Jitendra Malik, Ken Goldberg · · 2026

Summary

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,

Abstract Summary

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.

Key Points

Framework for learning bimanual policies from single-arm robot + human collaboration.
Collects single-arm demonstrations while human performs the other side, then swaps roles.
Uses hand pose estimation, segmentation, and inpainting to create synthetic bimanual demos.
Zero-shot deployment on unseen bimanual configurations; up to 70% success.
Few-shot fine-tuning with 25 target demos boosts performance by 65-70% over scratch.

Problem Statement

As described in Figure 1 , MonoDuo collects a demonstration dataset $\mathcal{D}^{\mathcal{S}}=\{\tau_{1}^{\mathcal{S}},\tau_{2}^{\mathcal{S}},...,\tau_{n}^{\mathcal{S}}\}$ consisting of $2N$ successful trajectories of a source robot-human pair $\mathcal{S}=(\mathcal{S}_{r},\mathcal{S}_{h})$ performing some task.

MonoDuo

In this section, we describe more details of how MonoDuo enables the learning of bimanual robot policies when only a single-arm robot is available. An overview is shown in Figure 1 . For each bimanual task, we collect a source dataset $\mathcal{D}^{\mathcal{S}}$ using a human to teleoperate a single-arm robot to collaborate with a human partner on the task.

To ensure a balanced data distribution, the roles of left-arm and right-arm are alternated across episodes.

MonoDuo — MonoDuo: Using One Robot Arm to Learn Bimanual Policies — Fig. 3 : Examples of zero-shot rollout on the target bimanual UR5e. Left: Lift Box; Right: Pack Bag. Single-Arm policies do not coordinate the actions well, leading to asynchronous movements as shown in the Lift Box task and collision in the Pack Bag task. Policies trained without cross-painting are less robust and misgrasps often. MonoDuo exhibits coordinated behaviors while being precise.

Experiments

We use a Franka arm as the single-arm source robot, and a pair of UR5e arms setup as the bimanual target robot. For RGB-D data collection, we use a ZED2 stationary fixed camera and a ZED-mini wrist-mounted camera.

Sources and Demonstrations

Figures are reproduced from Bajamahal et al. See the full paper for experimental details and the project page, when the paper links one, for demonstrations and videos.