PA-BiCoop: A Primary-Auxiliary Cooperative Framework for General Bimanual Manipulation
Qicheng Bai, Ziru Wang, Teli Ma, Guang Dai, Jingdong Wang, Mengmeng Wang
AI summary
Problem
Existing bimanual manipulation methods either rely on dual independent models that lack inter-arm knowledge sharing or use single shared models that treat both arms as functionally equivalent, ignoring the adaptive division of labor essential for complex coordination.
Approach
PA-BiCoop uses a single shared encoder with specialized primary and auxiliary decoders to predict actions, coupled with a learnable module that dynamically assigns left or right arms to primary or auxiliary roles based on task context.
Key results
- 48% average performance gain over SOTA in RLBench2 simulations
- Over 50% average improvement in real-world dual-arm tasks
- Automatic dynamic role assignment between left and right arms
- Enhanced inter-arm knowledge sharing via shared global features
Why it matters
Provides a scalable, role-aware architecture for dual-arm robots, enabling more efficient and flexible coordination in complex manipulation tasks for robotics researchers and automation engineers.
Abstract
Bimanual manipulation is essential for advanced robotic systems because it offers higher efficiency and flexibility compared to single-arm configurations. However, existing ap- proaches either lack inter-arm interaction or ignore the need for a dynamic division of labor, treating the arms as functionally equivalent. To address these limitations, this paper draws inspiration from human bimanual manipulation where one arm handles core operations and the other provides auxiliary support, and proposes PA-BiCoop, a new single-model biman- ual cooperation framework with dynamic primary-auxiliary arm differentiation. PA-BiCoop categorizes robotic arms into primary and auxiliary arms with adaptively adjustable roles across task stages, employs two specialized decoders that share a global feature encoder: the primary decoder generates the primary arm’s base-coordinate pose and core-task affordance heatmaps, and the auxiliary decoder outputs the auxiliary arm’s relative pose in the primary arm’s coordinate system. Moreover, we design a dynamic role assignment module to automatically map roles to left/right arms without manual pre-definition. This design facilitates inter-arm knowledge sharing and coordinated manipulation. Extensive experiments demonstrate that our PA- BiCoop achieves superior performance: it outperforms state-of- the-art baselines by 48% on average in RLBench2 simulation tasks and by over 50% on average in real-world tasks, thereby verifying its effectiveness and advancement in bimanual ma- nipulation.