Design of an intelligent feed forward controller system for vehicle obstacle avoidance using neural networks

The design of a novel feed forward controller system for vehicle obstacle avoidance using the neural network methodology is proposed. Currently, most obstacle avoidance systems are designed based on a segmented procedure: a) parametric path planning; b) desired yaw moment computation based on a simplified model; c) yaw moment tracking; d) stable controller design. In this paper, a different strategy is followed. An intelligent 'autopilot', that has been trained using a set of optimised obstacle avoidance manoeuvres, decides how to avoid the obstacle. The obstacle avoidance manoeuvres have been optimised using a reformulation of the Pontryagin's Maximum Principle and global numerical optimisation techniques. The proposed controller has the advantage that it respects 'by design' the internal dynamics of the system and can be adjusted in order to account any model uncertainties. Furthermore, it is computationally very efficient. The performance of the intelligent system is evaluated by means of simulations in MATLAB for a number of test cases.