Project information:
AIVATAR. Artificial Intelligence based modular Architecture Implementation and Validation for Autonomous Driving (October 2022 – September 2025). Reference: PID2021-126623OB-I00. Financed by the Spanish Ministry of Science and Innovation (MICIIN).
Abstract:
According to the United Nations (UN), more than two thirds of the world population will live in cities by 2050. Providing safer, more sustainable and efficient mobility for goods and people in populated cities is a priority identified in Horizon Europe and the Spanish Plan for Scientific, Technical and Innovation Research 2021-2023. The transformational change in mobility will contribute to the Sustainable Development Goals (SDG) for UNs 2030 Agenda, to the European Green Deal, as well as to the Vision Zero Deal. One key technology for the future mobility is Autonomous Driving (AD), which has been a hot topic in the last decade that has caught the attention of research and industry. There are numerous different fully autonomous vehicle projects in various stages of development. However, autonomous vehicle technology is far from being ready to be deployed at full scale. Connected, Cooperative and Automated Mobility (CCAM) European Partnership identifies four problems for the AD introduction in the market:
1) Insufficient demand.
2) AD techniques are not yet sufficiently mature.
3) Current R&I efforts are fragmented and lack a long-term vision.
4) Demonstration and scale-up is limited.
On the other hand, the vast amount of data that our society is generating every day (Big data) and the exponential increase in computational power have revolutionized the Artificial Intelligence (AI) field, boosting Machine Learning (ML) and Deep Learning (DL) paradigms. The recent DL algorithms are extremely powerful tools that can contribute to solve the afore mentioned problems.
Our group has been working in the Tech4AgeCar project (2019-2021), developing classical technologies that enable our open-source automated car with autonomous driving capabilities, which have been tested in simulation in the challenging CARLA Leaderboard and in real field tests in restricted uses-cases. Based on the results in our former project, AIVATAR proposes a robust and modular human-like Autonomous Driving architecture that evolves our current techniques by cloning human behaviors through deep learning techniques, with the goal of achieving safe and sustainable navigation in challenging scenarios in both simulation and controlled real conditions, paying special attention to the validation methodology and the explainability of the decisions to achieve user-acceptance.
The proposal is disruptive and present five main breakthroughs:
1) Implement a long-term trajectory prediction with multi-head attention to find out the intentions of the surrounding objects (vehicles and VRUs) to be able to make decisions in advance as humans do.
2) Address a complete planning system, based on a hybrid structure, with a high-level decision-making layer based on Hierarchical Multiagent DRL and maneuver control based on Model Predictive Control (MPC) with the goal to manage complex traffic decisions imitating human behaviors.
3) Design a vehicle to driver (V2D) communication protocol based on questioning/answering interaction using natural language and automatic speech recognition/synthesis, contributing to improve explainability of the AD decisions.
4) Provide a modular AD architecture able to be tested in simulation and in real conditions.
5) Propose standard validation methodology that allows to evaluate the modules and the whole architecture in a holistic way in simulation, minimizing the gap between simulation and real-world evaluation.
Research Team:
Bergasa Pascual, Luis M.
Barea Navarro, Rafael
Revenga de Toro, Pedro
Ocaña Miguel, Manuel
López Guillén, Elena
Escudero Hernanz, Marisol
García Garrido, Miguel Ángel
Llamazares Llamazares, Ángel
Work Team:
Juan Felipe Arango Vargas
Carlos Gómez Huélamo
Javier Araluce Ruiz
Rodrigo Gutiérrez Moreno
Miguel Antunes García
Santiago Montiel Marín
Pablo Pardo Decimavilla
Ricardo Ignacio Pizarro Carreño
Fabio Sánchez García
