Dynamical Obstacle Avoidance of Task- Constrained Mobile Manipulation Using Model Predictive Control

Abstract

Task-constrained motion planning (TCMP) is involved in many practical applications, such as opening the door, opening the drawer, twisting the screw, and so on. Because of the trend of man–machine collaboration and the existence of dynamic in environments, planning the collision-free motions for task-constrained manipulation is a significative problem. This paper explores the TCMP issue for mobile manipulation, which uses a mobile base to enhance the working range and flexibility but simultaneously makes the problem harder than that of single manipulation because of the additional degrees of freedom (dofs). We propose an optimization-based method to plan the obstacle avoidance motion in real-time. First, the global robotic Jacobian matrix that combines the omnidirectional base and the robotic arm is derived. Second, model predictive control is used to plan the control rule in order to maximize the closest distance between the obstacles and the mobile manipulator and minimize the velocities of null space at the same time. We have deduced the model with four differential equations that represent the law of the distance over time. Third, the distances are calculated and sent to the model to calculate the velocity of each joint of the arm and the base using ACADO that is an open-source toolkit. Using the velocities, the mobile manipulator can move away from the approaching people while still fixing the end of the arm to manipulate tool at the same time. Our method is verified on the mobile manipulator that consists of four Mecanum wheels, a base, a UR10 arm, and four kinect cameras in Gazebo simulation with using the ROS operation.