The main objective of the project is to develop new approaches and control architectures for vehicles to design shared control systems in a generic Human-centered perspective.
Numerous works in the autonomous driving field have shown that, defining behaviors adapted to all users (including Person with Reduced Mobility – PRM) and all situations that may be encountered is very complex, if not impossible (Vijay R. et al, 2021).
Two solutions can come at hand: either trying and having time to cover a large number of users and a large set of situations, being in a sense exhaustive for the target applications or, providing cooperation and especially giving the machines learning capabilities, with a goal to allow them to learn-and-adapt to users and situations. This research project clearly explores the second solution.
The solution developed necessitates two important fields – robust control and AI-learning – and the core will be to combine them in a framework that preserves real-time safety and performances. It does enter in the hot topics of the future of control as stated by (Recht 2019): "One final important problem, which might be the most daunting of all, is how machines should learn when humans are in the loop. What can humans who are interacting with the robots do and how can we model human actions?".
Project start date: January 31, 2022.
Duration: 36 months.
(*) Recht B. (2019). A tour of reinforcement learning: The view from continuous control. Annual Review of Control, Robotics, and Autonomous Systems 2, 253-279.
Project funded by the ANR within the framework of the 2021 generic call for projects