Biomimetic Control Methods for Nonholonomic Mobile Robots

Biomimetics can be expressed as biologically inspired mechanical devices. The term of itself is derived from bios, meaning life, and mimesis, meaning to imitate. This thesis deals with biomimetic approaches for controlling nonholonomic mobile robot. It obtains three techniques of such biomimetic approaches. They include adaptive actor-critic algorithms, feedforward and feedback techniques. The adaptive actor-critic algorithm is emphasized on a multi-step method of simulated experience through the predictive model. It provides m-predicted state; particularly, we investigate a learning method where the only measurement at time k is available so that it can provide m-step discounted rewards through the predictive model. The weights of actor and critic in the present approach are updated every sampling time. This concept is similar to the predictive control in the field of conventional control theory. The second method of biomimetic approaches is an NN-based feedforward technique, in which it is designed as a man-machine interface that is called a neurointerface. Such a method is useful to help a non-expert operator for handling difficult tasks. It changes the operational space through an NN, allowing the human operator to interact with the process through less-specialized commands. A word of neurointerface is chosen to emphasize the use of NN for the solution of interface problem. In the simulation the NN-based inverse model shows good results on mapping techniques. The last method is a feedback compensator of neurointerface. In order to reduce the effect of mapping errors, or the efect of external and internal disturbances, the feedback compensator is proposed. A PD control or an adaptive fuzzy compensator is implemented together with the feedforward NN in a parallel manner. It is shown that good results are obtained in an application of controlling a nonholonomic mobile robot with two independent driving wheels.