Hybrid neural network and genetic algorithm combination for recognition and avoidance of vehicle obstacles

Abstract

This paper is devoted to the research and application of a hybrid combination of neural networks and genetic algorithms for the recognition and avoidance of vehicle obstacles. The paper analyzes the current state of the problem of obstacle recognition by self-driving vehicles, the level of training of relevant systems for self-driving, and what risks and dangers it poses for road users. Various variants of recurrent neural network architectures are investigated, their recognition features are analyzed, and the main advantages and disadvantages are determined. A genetic algorithm is applied to determine the optimal parameters of the system for recognition and avoidance of vehicle obstacles. In the developed intelligent system, the mechanics of moving the car were implemented, the road and lanes were created, artificial sensors were designed to detect obstacles, traffic simulation was implemented, a neural network and a genetic algorithm were designed and implemented, and parallelization of objects was carried out to simulate all possible options for avoiding obstacles. Testing the developed system for recognition and avoidance of vehicle obstacles showed its high efficiency when combining a neural network and a genetic algorithm.