Published on: 2026-09-11
Source: Saint Petersburg State University of Architecture and Civil Engineering –
An important disclaimer is at the bottom of this article.
A scientist from SPbGASU studied how reinforcement deep learning technologies and simulation modeling can be used to optimize the operation of autonomous taxis in an urban road network. The developed models allow for consideration of changing passenger flow, the number of vehicles on the roads, and their speed.
The study was conducted by a Doctor of Technical Sciences, a professor at the Department of Information Technology and Mathematical Modeling. Alexey Namestnikov. The work is dedicated to the integration of reinforcement learning models with simulation models using the example of autonomous taxi behavior in urban traffic conditions.
Alexey Namestnikov
The development of unmanned transport technologies opens up the possibility of using autonomous vehicles in urban public transport systems. At the same time, one of the tasks is to ensure the accessibility of such transport for passengers: unmanned taxis must be located where there is the greatest demand for them. This demand, in turn, can change depending on passenger flow intensity and road conditions.
To address this task, the study utilized the AnyLogic simulation modeling tools. Based on them, hybrid models were developed that combine discrete-event and agent-based approaches to simulating a fleet of autonomous taxis as integrated intelligent systems. These models were supplemented with reinforcement deep learning technologies.
This approach makes it possible to model the behavior of driverless taxis taking into account changing conditions of the urban transport system and to select the optimal placement of vehicles in the road network. As a result, it becomes possible to simultaneously consider several factors – changes in passenger flow, the number of vehicles, and the speed of urban traffic.
The scientific novelty of the project lies in the development of a new method for the optimal placement of unmanned taxis in the urban road network, taking into account stochastic demand. Unlike known approaches, the proposed method is based on the integration of simulation models and deep reinforcement learning models.
In addition, within the study, a new deep reinforcement learning model called Deep-Q-Network (DQN) was developed. Its distinctive feature is a neural network architecture adapted to the problem being solved.
“The development of multi-agent deep reinforcement learning systems allows for more complex modeling of unmanned vehicle behavior. A promising direction is their training based on simulation modeling using geoinformation data,” noted Alexey Namestnikov.
Alexey Namestnikov has been conducting research in the field of applying intelligent technologies in the design of technical systems since 1997. The scientist has published about 120 scientific papers on this subject. In the future, the scientist plans to continue working on the development of multi-agent deep reinforcement learning systems for training the behavior policy of unmanned vehicles. Simulation modeling and geoinformation data are intended to be used as the basis.
The research was carried out within the framework of a grant for scientific research work by the scientific and teaching staff of SPbGASU in 2026.
Please be advised; This information is raw content obtained directly from the information source. It is an exact report of what the source claims and does not necessarily reflect the position of MIL-OSI or its clients.