Fuzzy Controller for Autonomous Vehicle Based on Rough Sets

This article presents a fuzzy controller for autonomous vehicle to intelligently recognize running environment and avoid an obstacle, which is constructed by rough sets (RSs) and an adaptive neuro-fuzzy inference system (ANFIS). Firstly, RSs are considered to propose a pyramid normalization (PN) method for normalizing state parameters (SPs) which are defined to recognize relative position such as distance and angle among a vehicle, an obstacle and target pathway, to improve the adaptability of complex environment and optimize the database of driving knowledge. Secondly, ANFIS is employed to design a controller with self-position azimuth correction (SPAC) for performing trajectory tracking and obstacle avoidance. Finally, the proposed methods have been implemented on the model vehicle called “RoboCar”, and compared with various fuzzy control approaches such as the initial SPs with ANFIS, the normalized SPs with fuzzy neural network (FNN), and the initial SPs with FNN. Time, maximum tracking error and mean tracking error are calculated to evaluate the performance. The experimental results with four kinds of target pathways have shown that the PN-ANFIS-based controller has saved time (7.6%), reduced maximum tracking error (8.1%) and mean tracking error (8.5%).