Development of a frontal collision avoidance system using a neuroevolutionary approach

D. D. Yaparov^a, S. V. Bessmertnyi^b, Iu. D. Danilova<sup>b

a</sup> South Ural State University (National Research University), Chelyabinsk, Russian Federation
^b Moscow Institute of Physics and Technology, Moscow, Russian Federation

Abstract: We developed a frontal collision avoidance system based on artificial neural networks. We trained the network using evolutionary algorithms, which optimized the network structure and hyperparameters, including activation functions and the number of hidden layers. We evaluated model quality using the loss function and data processing time.
During computational experiments, we selected the best-performing models and verified them on real-world data with a car under various driver response scenarios. The test results showed that the proposed model achieves high accuracy and efficiency, both on training data and in real operating conditions, making it a promising solution for enhancing road safety.

Keywords: neural networks, information processing, frontal collision avoidance systems, evolutionary algorithms, learning strategies.