A Behavior-based Adaptive Dynamic Programming Method for Multiple Mobile Manipulators Coordination Control

In this work, a behavior-based adaptive dynamic programming (BADP) method is proposed for coordination control of unmanned ground vehicle-manipulator systems (UGVMs). Through a null-space-based behavioral control (NSBC) framework, the multi-objective coordination control is transformed into a single-objective tracking control at the mission layer. Since cost functions and control constraints are simplified at control layer, the complexity of controller design is reduced. Then, an identifier-actor-critic reinforcement learning algorithm framework is introduced to learn the optimal control policy by balancing the control performance and consumption. Simulation results show that control costs are reduced around 13.5% per sampling period compared to existing multiple objective control methods. Finally, the BADP method is experimentally validated using four real UGVMs.