Predictive Control Method Based Robot Wheeled Formation

Nowadays,as the development of the robot technologies,the robotic research has attracted more and more attention. The robot can be applied in many fields instead of human, which can save the human resources and reduce the risks. Among various kinds of robots, the wheeled mobile robot is a typical representation. However, as increasing of the work complexity, one single wheeled mobile robot cannot fulfill the requirement of some specific tasks, so the formation control include multiple wheeled mobile robots has become a new popular topic in robotic research.This work began with the single robot’s control, formulated the systems of the robot stabilization and trajectory tracking model based on the kinematic model of nonholonomic constrained mobile robot. In order to control these two systems, the Model Predictive Control(MPC) method was applied in the experiments. Through formulating a constrained Quadratic Programming(QP) optimization problem, the MPC method can obtain the optimal control input by solving this optimization iteratively.Then, based on the research of single wheeled robot, we developed a leader-follower framework for the mobile robot formation, then a Separation-Bearing-Orientation Scheme(SBOS) formation system model was formulated by using geometric relationships. The system states include the distance between leader and follower、bearing、and the error of orientation. Meanwhile, in order to avoid follower colliding with the obstacles in its way, we developed a Separation-Distance Scheme(SDS) obstacles avoidance system model. Two systems can be switched during the formation and the system can be control to reach the desired robots’ relationships.For the QP problem, we proposed the Primal-Dual Neural Network(PDNN) which can solve the constrained QP problem formulated by MPC and obtain the optimal solutions efficiently. On the other hand, a collective optimization method, which integrated Particle Swarm Optimization(PSO) with General Projection Network(GPN), was applied for solving QP problem, this method can search the global optimal solutions with high efficiency.In this work, the MPC based on two optimization method was implemented on the robot stabilization, trajectory tracking and robot formation, and four experiments were performed to verify the effectiveness of the proposed approach.