Lifelong learning for UAVs

3L4AV is a mobility research project between the Czech Technical University in Prague (CTU) in Czechia and the University of Technology of Belfort-Montbéliard (UTBM) in France. The research goal is to provide new methods for autonomous vehicles to improve the robustness of their perception system, in particular for dynamic objects detection and tracking in adverse conditions, such as during adverse weather and dense traffic.

In this project, we will investigate machine learning methods for multisensor systems deployed in autonomous vehicles. The heterogeneous nature of the sensory data will allow mutual training of the methods of dynamic object detection and tracking. On-the-fly, lifelong learning of the objects models for detection and tracking will be achieved through exploitation of the heterogeneity and amounts of data gathered by the vehicle’s sensors over long periods of time. In contrast to existing technologies which are mainly based on static models, our project will focus on development of adaptive models that are completed and refined based on the data gathered over long-time operation of the autonomous vehicle.