AI4RMoSer
Artificial Intelligence for road mobility service
PID2021-124335OB-C21
Proyectos de Generación de Conocimiento

The use of autonomous vehicles (AVs) will be a reality in the coming decades, although to achieve this the underlying technologies must reach a degree of maturity that they do not yet have. Regarding vehicles driven by humans, ten Advanced Driving Assistance Systems (ADAS) will be mandatory in Europe on all new cars from July 2022. AVs and ADASs use analogous perception technologies leading to important synergies. Finally, the advantages of electric motors over internal combustion engines are becoming increasingly apparent. Among other aspects, the management of vehicle batteries and Vehicle to Grid (V2G) technology that allows energy to be returned to the electricity grid from the battery of an electric car, will be important.
In recent years, Artificial Intelligence (AI) has had exponential growth in many fields. The three fields described above are no exception. A recent McKinsey Analytics survey ranks the automotive, transportation and electric power sectors in the top ten in AI capabilities. Deep Learning (DL) has decisively influenced the analysis of sensory information. But for the AVs to be able to move around the environment in a safe way, it is not enough to simply classify the objects around them, it is necessary to predict what their trajectory will be in the coming moments; this is one of the objectives of this project. Likewise, the use of the DL for driver monitoring is of vital importance. An ADAS is proposed to prevent dangerous situations based on the fusion of information from the driver, the vehicle, and its environment.

In addition, it is wanted to improve the user experience of the driver and passengers taking into account their expectations. To this end, the use of the paradigm of intelligent agents, fuzzy logic, as well as machine learning techniques is proposed. Additionally, AI is beginning to be successfully applied for energy management in vehicles and in V2G. Intelligent agents and agent-based simulation allow the integration of different ADAS and the simulation of EAV fleets including their management and integration with the environment. In addition, it is required to demonstrate the potential of VAEs to provide energy flexibility services to distribution network operators through energy consumption (G2V or network to vehicle) or energy injection (V2G) as needed. The charging and discharge process should be coordinated with low-power renewable energy sources, taking into account consumer demand and the variability of energy prices. The “energy as a service” framework for VAEs will combine V2G service for energy exchange and energy management between the smart grid and the autonomous vehicle. Self-awareness techniques can detect when an anomalous situation is occurring that indicates that the vehicle has made a wrong decision or that the driver is not driving as usual. The project will study a new methodology based on a probabilistic switching model and a bank of Adversarial Generative Neural Networks.
Because of the type of task EAVs and ADAS are, and the level of trust they require, IA-based systems must have the ability to explain their decisions. Therefore, the project also emphasizes a trustworthy IA.
