This paper presents a novel navigation strategy of robot to achieve reaching target and obstacle avoidance in unknown dynamic environment. Considering possible generation of uncertainty, disturbances brought to system are separated into two parts, i.e., bounded part and unbounded part. A dual-layer closed-loop control system is then designed to deal with two kinds of disturbances, respectively. In order to realize global optimization of navigation, recurrent fuzzy neural network is used to predict optimal motion of robot for its ability of processing nonlinearity and learning. Extended Kalman filter method is used to train RFNN online. Moving horizon technique is used for RFNN motion planner to guarantee optimization in dynamic environment. Then, model predictive control is designed against bounded disturbances to drive robot to track predicted trajectories and limit robot’s position in a tube with good robustness. A novel iterative online learning method is also proposed to estimate intrinsic error of system using online data that makes system adaptive. Feasibility and stability of proposed method are analyzed. By examining our navigation method on mobile robot, effectiveness is proved in both simulation and hardware experiments. Robustness and optimization of proposed navigation method can be guaranteed in dynamic environment.
Robust Optimal Navigation Using Nonlinear Model Predictive Control Method Combined with Recurrent Fuzzy Neural Network
