The Department of Mechanical Engineering of Politecnico di Milano is one of the partners involved in the European Project AI@EDGE which aims to virtually validate the behavior of cooperative connected and automated vehicles. The general aim of the project is to develop advanced technologies for 5G networks. The focus is on how such networks can be improved by using Artificial Intelligence (AI) and edge computing. The research is devoted to understand how to apply AI and edge computing to road traffic management. The flow of Cooperative, Connected and Automated Vehicles (CCAVs) into a roundabout is optimized. At the moment, this kind of infrastructure represents a bottleneck for automated vehicles. Such vehicles often get stuck into roundabouts because they are not able to keep accident risk at the needed minimum level. The negotiation of roundabouts by CCAVs could lead to several benefits, such as safer traffic (reduced number of accidents), increased traffic fluidity and less air pollution due to a decrease in the stop-and-go events at the junction. To achieve these goals, a new approach based on 5G networks and edge computing has been developed. A MEC/Edge node receives data about the states of the connected vehicles (position and velocity) for a given time instant, elaborates them and calculates the states of the automated cars for the following time instant. This information is sent back to the CCAVs, so that they can move in the right direction and at the computed velocity. This approach has been called “Vehicle-to-Network-to-Vehicle” (V2N2V). The calculations are performed by an AI algorithm, trained by means of a Reinforcement Learning (RL) process. The RL policy is trained with the objective to minimize the time needed for automated cars to go though the roundabout, without forgetting the comfort of the passengers. In fact, the longitudinal and the lateral accelerations are constrained to a level which is acceptable by the passengers of the vehicles.
In order to test the algorithm which controls the behaviour of the CCAVs, the driving simulator of Politecnico di Milano (DRiSMi laboratory) has been used. The research group of the Mechanical Engineering Department is led by Prof. Gianpiero Mastinu, which collaborates with Prof. Massimiliano Gobbi and Prof. Giorgio Previati.
The use of a dynamic driving simulator in this research represents a great opportunity to test the AI policy in a safe environment, avoiding the risks connected to real traffic and the possibility of congestions and accidents. Moreover using the driving simulator it is possible to explore different traffic situations simply modifying the parameters of the simulation, thus reducing the costs of experimental set-up. Finally, the use of a driving simulator ensures the repeatability of the test, eliminating the influence of disturbing factors such as adverse weather conditions, which may reduce drivers’ visibility. The main objective of the tests is to understand if the behaviour of automated cars can be easily accepted by human drivers or if it is perceived as somehow unnatural. So, a digital twin of an actual four-leg roundabout has been reproduced on the driving simulator. During the tests, the traffic is constituted by both simulated human-driven vehicles and simulated CCAVs. Drivers in the driving simulator are asked to approach the roundabout, go through the circulatory roadway and then leave it to the third exit. In this way, drivers can interact with entering flows of vehicles coming from other legs of the roundabout. The tests are repeated with different percentages of automated vehicles involved in the simulations. After that, drivers are asked to fill in a questionnaire, commenting their perceptions on traffic smoothness and their safety feeling. Most of the drivers involved in the tests had positive impressions on the simulated traffic situations, not highlighting any disturbing behaviour by automated vehicles. On the contrary, often they felt safer as the amount of CCAVs in the scenario increased. This gives encouraging insights on the possibility for automated vehicles to negotiate roundabouts and on the acceptance of their behaviour by “traditional” human drivers.
AI@EDGE project will be completed in December 2023, so during the next moths, the final tests and the final validation of the system will be performed. The outcome of the project will represent a step forward in the direction of the everyday use of cooperative, connected and automated vehicles on the streets.
The full list of the partners of the AI@EDGE Consortium can be found at aiatedge.eu.
To find out more about the interaction among Cooperative, Connected and Automated Vehicles and vehicles with human drivers, join the event part of the series “Visioni Politecniche” entitled "Mobilità cooperativa, connessa, automatizzata e sostenibile" that will be held during the third edition of Festival dell’Ingegneria from the 22nd to the 24th of September 2023.