Advances in Fuzzy Logic Control for Lateral Vehicle Guidance

A candidate for intelligent control for lateral guidance of a vehicle is a fuzzy logic controller (FLC). The details of the most recent FLC with a rule base derived from heuristics and designer insight to the problem at hand, as well as a method for estimating a projected lateral displacement are presented along with simulation results. The structure of this FLC is suited to incorporate human knowledge about the steering operation by its choice of inputs and outputs of the FLC which are natural to human steering operation. In addition, a method that makes use of preview information regarding upcoming road curvature is developed and simulated based on human steering operation. This method projects an estimate of the lateral error relative to the reference track at a specified look-ahead time. Simulations imply that there exists an optimal look-ahead time. Too short a time does not make full use of the preview information, while too long a time results in inaccurate estimates of future lateral error, degrading the performance. Success of the FLC strongly depends on the set of rules provided by the designer. It is extremely important to have methodologies for rule development. Therefore, a methodology is considered for automatically generating a fuzzy rule base controller, specifically, a Neural-Network Driven Fuzzy Logic Controller (NNDFLC). This process automates the design of control parameters and uses a clustering technique to cluster the space of inputs to the controller in order to take advantage of local characteristics in the reasoning of the control outputs. Given enough input/output data to train the neural networks involved, simulations show its success. However, this method requires the generation of a complete set of training data, covering the entire space of inputs to the controller. This is an important factor in the design of the NNDFLC with regard to the convergence of control parameters in the learning process and for taking advantage of the clustering features of the controller. Table of