Autonomous vehicle obstacle avoiding and goal position reaching by virtual obstacle

Abstract

The problem of dynamic path generation for the autonomous vehicle in environments with unmoving obstacles is presented. Generally, the problem is known in the literature as the vehicle motion planning. In this paper the behavioural cloning approach is applied to design the vehicle controller and virtual obstacle is used also in the goal position reaching. In behavioural cloning, the system learns from control traces of a human operator. To learn from control traces the machine learning algorithm and neural network algorithms are used. The goal is to find the controller for the autonomous vehicle motion planning in situation with infinite number of obstacles.