Nonlinear Model Predictive Control for Mobile Medical Robot Using Neural Optimization

Abstract

Mobile medical robots have been widely used in various structured scenarios, such as hospital drug delivery, public area disinfection, and medical examinations. Considering the challenge of environment modeling and controller design, how to achieve the information from the human demonstration in a structured environment directly arouse our interests. Learning skills is a powerful way that can reduce the complexity of algorithm in searching space. This is especially true when naturally acquiring new skills, as mobile medical robot must learn from the interaction with a human being or the environment with limited programming effort. In this article, a learning scheme with nonlinear model predictive control (NMPC) is proposed for mobile robot path tracking. The learning-by-imitation system consists of two levels of hierarchy: in the first level, a multivirtual spring-dampers system is presented for imitation of the mobile robot's trajectories; and in the second level, the NMPC method is used in the motion control system. The NMPC strategy utilizes a varying-parameter one-layer projection neural network to solve an online quadratic programming optimization via iteration over a limited receding horizon. The proposed algorithm is evaluated on a mobile medical robot with an emulated trajectory in simulation and three scenarios used in the experiment.