Fast neural network control of a pseudo-driven wheel on deformable terrain

The performance of the wheel-driving control scheme in wheeled mobile robots is a vital aspect in vehicular robotics applications, both for fully exerting the robot tractive capabilities and to save energy. This is especially true in contexts in which the robot must travel through unknown and unpredictable deformable terrains, such as the planetary exploration context considered in this study. To compensate the disturbances resulting from terrain deformations while simultaneously leveraging the control advantages provided by the use of the recently proposed concept of pseudo-driven wheels (PDWs), an artificial-neural-network-based control method is proposed here. This study develops and presents the network algorithms necessary for achieving active following control on velocity tracking for PDWs. To handle the considered complex and highly uncertain wheel-terrain interactions, an online sequential forgetting update method for the neural network is presented, and an improved online sequential extreme learning machine (OS-ELM), combined with a proportional integral derivative (PID) controller is used, thereby leading to an efficient and highly performant hybrid OS-ELM-PID control system. A simulation showed the feasibility of the proposed control method, and subsequent real-life experimental results demonstrate the capability of the control system in maintaining the drawbar force on the PDW within the range F_DP = ±2 N.