| |
At the present stage of social development, accompanied by global digitalization, driver assistance systems have become particularly relevant when driving, which allow monitoring the traffic situation with the help of artificial intelligence. Such systems have certainly made modern human-driven vehicles more efficient and safer. At the same time, the technology of unmanned vehicles (cars), despite the availability of appropriate technologies, has not yet become an everyday reality. This practical complexity is due to the fact that a new generation of unmanned systems needs algorithms capable of preventing traffic accidents and coping with complex traffic situations, just as a person does while driving a vehicle, taking into account an optimally complete list of intersection paths by various road users.
Based on this, it is necessary to develop such a model of neural network training that will allow networks to learn in real time, tactically building scenarios for crossing intersections in real time, that is, building a model for solving complex multi-purpose tasks. At the same time, to solve the modeling problem, it is not enough to simply develop a network model and scenario options for its training - in order to properly train an artificial neural network, it is necessary to choose an appropriate learning algorithm that will solve the task in an optimal way. Of the available learning algorithms, the most optimal seems to be the algorithm of a deeply deterministic policy gradient (DDPG). The specifics of this algorithm will ensure the stability and efficiency of the model. The Deep deterministic Policy Gradient (DDPG) algorithm combines the advantages of the Subject-Critic and DQN ((Deep Q-Learning) algorithms, choosing the best of them and thereby solving the complex problem of the continuous action space through experience reproduction and asynchronous updating, which becomes particularly relevant in those systems that are associated with unmanned driving vehicles.
Keywords:genetic algorithms, optimization problem, artificial neural networks, neural network learning algorithms, deeply deterministic policy gradient algorithm.
|
LEGAL
INFORMATION:
