Distributed Safety-Aware Motion Planning for Dual-Arm Systems via Gaussian Belief Propagation With Application to Catenary Dropper Assembly

Abstract

With the increasing deployment of dual-arm robots in compact and shared workspaces, generating safety-aware and efficient motion plans for such high-degree-of-freedom (DoF) collaborative systems remains a challenge. Centralized planners suffer from severe computational bottlenecks, while existing distributed optimization methods are highly sensitive to initialization in high-dimensional non-convex spaces and easily trapped in local minima. This letter presents a distributed motion planning method for dual-arm systems based on Gaussian Belief Propagation (GBP). We formulate the trajectory optimization problem as a probabilistic factor graph and solve it through parallel Gaussian message passing over discrete waypoint nodes. The method treats each manipulator as an independent inference agent and exchanges only neighbor beliefs at coupled waypoints, making it extensible to larger multi-arm settings under locally bounded interaction graphs. To validate the effectiveness of the proposed method, we benchmark it against a centralized factor-graph solver and advanced distributed optimization-based planners in batch dynamic simulations with four manipulators, and further apply it to a highly coupled and confined task of dual-arm catenary dropper assembly. Results show that the proposed planner achieves a 94.25% success rate, the shortest per-goal trajectory length and completion time, and a 0.081 m mean safety clearance under the 0.1 m margin setting in four-manipulator dynamic-obstacle simulations, and maintains real-time executable behavior in the dual-arm catenary-dropper experiment, demonstrating practicality for scalable multi-manipulator planning and safety-aware assembly.