Exponential navigation functions with a learning algorithm

This paper suggests a method for autonomous wheeled mobile robots navigation under the nonholonomic constraint. The suggested method uses navigation functions that are based on the polar kinematics equations, where the steering angle and the orientation angle of the robot are included in an exponential function of the line of sight angle. Another control law is suggested for the robot's linear velocity to drive the robot to a desired position with a desired final orientation angle. The exponential navigation functions depend on various navigation parameters that allow to change the robot's path. This approach is combined with the collision cone technique to avoid collision. A Q-learning algorithm is suggested to select automatically the appropriate values of the navigation parameters. Simulation is used to illustrate the method.