Application of machine learning for NonHolonomic mobile robot trajectory controlling

Mobile robots are very interested by researchers over the last few years because of their applications and physical characteristics. The workspace of mobile robots is not always ideal, but typically filled with disturbances (known or unknown) such as uneven surface terrain, natural friction, uncertainties, and parametric changes. In this study, a new approach namely active force control (AFC) scheme integrating artificial neural network (ANN) has been suggested to cope on the disturbances and thus improve the trajectory tracking characteristic of the system. Therefore, a two wheeled mobile robot has been simulated, and ANN technique is explicitly employed for the estimation of the inertia matrix that is needed in the inner feedback control loop of the AFC scheme. The robustness and efficiency of the identified control scheme are studied considering various forms of loading and operating conditions. For the purpose of benchmarking, the AFC scheme performance has been compared to PID controller.