Hybrid Learning-Optimization Control Methods for Dual-Arm Robots in Cooperative Transportation Tasks

Abstract

This article presents a hybrid learning-optimization control method (HLOCM) for dual-arm collaboration. A model predictive optimized dynamic movement primitives (MPO-DMPs) framework is proposed for path planning of robotic arms, which utilizes DMPs to model demonstrated trajectories, while integrating model predictive control (MPC) to generate additional control inputs for local path planning in the presence of obstacles. This ensures smooth and adaptive motion adjustments without compromising trajectory stability. In the context of dual-arm cooperative transportation tasks, the leader arm is teleoperated, while the follower arm collaborates using admittance control to dynamically adjust its motion based on external forces and environmental constraints. By combining skill learning with compliant control, this approach not only enables precise and adaptive motion execution but also enhances the robot’s ability to generalize learned skills across different environments. Moreover, by integrating obstacle avoidance within the skill learning framework, the proposed method extends the applicability of learned robotic skills to dynamic and unstructured environments, making it well-suited for real-world manipulation and transportation tasks. Experiments are conducted on the Baxter robotic platform to validate the proposed algorithm.