A Dynamic Gradient Modulated Barrier Neural Network for Collision Avoidance of Mobile Dual-Manipulator Humanoid Robot

In order to generate collision-free motion of two redundant manipulators for mobile humanoid robots under complex constraints, a novel dynamic gradient modulated barrier neural network (DGM-BNN) is proposed to track end-effector path accurately and prevent potential collisions. For example, the mutual collisions between two manipulators and the collisions between manipulators and obstacles. The kinematic model of the mobile dual-manipulator humanoid robot is first established, and the trajectory tracking problem is formulated as a timevarying quadratic programming (TVQP) problem. Then, collision avoidance constraints are incorporated into the TVQP problem as inequality constraints through a shortest-distance-based angular velocity control strategy. The proposed Karush-Kuhn- Tucker (KKT) conditions with dynamic gradient modulation can overcome the possibility of convergence failure of the traditional KKT method in solving the optimal solution. Subsequently, A DGM-BNN solver model is designed based on neural dynamics formula. The excellent performance of the DGM-BNN scheme for reducing the position error and improving the collision avoidance capability is verified by simulation experiments. The results show that the mobile dual-manipulator humanoid robot can achieve higher-accuracy trajectory planning and collision avoidance than other methods, even under complex constraints.