The aim of the CAVSense project is to study the feasibility of one or more relocalization techniques on a proof of concept level. More specifically, the project is targeting towards exploring the feasibility to implement ego-vehicle localization in a given map of the environment based on the output of a multitude of sensors installed in the car (cameras, GPS, Sonars). Therefore, the main task is to develop a proof of concept that reaches the desired accuracy of the state-of-the-art (as it has been derived in Academic benchmarks), by
-
Detecting a set of localization features which are memory efficient and reusable for other tasks such as planning and behavior generation
-
Generating an online built-up local map to resolve ambiguity when associating detections with map elements
-
Implementing a novel pose graph adjuster for robust localization based on asynchronous and delayed measurements