Studies on lateral control and lane changing algorithms for application in autonomous vehicles

Abstract

The studies reported in this thesis focus on specific problems in the control of autonomous vehicles in a mixed traffic scenario whereby the road is shared by autonomous driving machines as well as human driven vehicles. Among the major concerns are the lateral and the lane change problems in Automated Highway Systems (AHS) and mixed mode traffic settings. The studies in this thesis also contribute to theoretical knowledge in the area of soft computing. Firstly, a novel fused Neural Network (NN) controller based on task decomposition is proposed. The proposed NN controller structure has been applied to a class of benchmark systems that require two input variables such as displacement and orientation in order to demonstrate its effectiveness. It has been tested for lateral control of autonomous vehicles under simulated and experimental environments. Secondly, an innovative encoding scheme coined as Fire Rules Chromosome (FRC) encoding scheme is proposed which can improve the convergence speed of a fuzzy controller optimized by Genetic Algorithms. Although it is a general purpose controller and can be applied to a variety of systems, here it has been developed as an efficient controller for lateral control. The robustness of this controller is studied by Monte-Carlo simulations. Lane keeping and lane changing are the two tasks involved in vehicle lateral control. The aim of the lane keeping is to maintain the vehicle at the center of the road and also follow the reference lane. The simplified neural network controller and fuzzy controller optimized by the proposed FRC scheme are implemented on a scaled vehicle for vehicle lateral control. For lane changing, the concept of virtual curvature is proposed to assist an automatic lane change maneuver. The virtual curvature algorithm incorporates virtual road curvature with bicycle model for vehicle lane change guidance. The virtual road curvature, does not physically exist, is a user assigned radius of a curved lane changing path which connects the current lane to the adjacent lane. The lane changing path guidance is achieved by assigning a virtual road curvature to the bicycle model to transform existing physical reference path curvature to the desired lane change path curvature. This transforming effect is accomplished by the inherent property of the bicycle model. The method is inspired by the observation that any change in the road curvature affects the vehicle lateral dynamics. The lane change maneuver offers flexibility in vehicle navigation, coordination and obstacle avoidance. However, during the lane changing, the merging vehicle should cross lanes which imply that the vehicle should consider obstacles or vehicles on adjacent lanes - a situation that could lead to accidents if not properly handled. The problem of lane change abortion is also studied through a computation of a maximum lateral displacement required for the lane change abortion. Collision free abortion point is defined in the study of lane change abortion. The maximum allowable lateral acceleration and the vehicle speed are the two prime factors governing the collision free abortion point. The effect of the two factors stated above on the abortion point is discussed in the thesis. A scaled prototype semi-autonomous vehicle is constructed for experimental testing of suggested algorithms in the thesis. The scaled vehicle is a modified Radio Controller (RC) car which is driving and steering by the front wheels. Infrared and ultrasonic sensors are installed to measure distance, and encoder sensor to measure vehicle speed. An industrial computer combined with A/D card is mounted on the vehicle as the main control system. The vehicle lateral model is obtained and it is verified that the scaled vehicle dynamic model exhibits the same properties with standard vehicle model.