Submitted Abstract
In the next decade, the increased number of assistive systems and the advent of semi-autonomous vehicles are expected to lead to a reduction of both the frequency and magnitude of road accidents, which will force insurance companies to adapt their offering. One option is to use telematics to monitor drivers and reward safe-driving habits by lowering premiums. Previous attempts have generally limited themselves to using sensors that are readily available on customers’ smartphones, such as GPS, but the collection of such data cannot capture the complexity of real-world driving scenarios.In this project, we collaborate with Foyer Assurances SA to research novel real-time risk assessment techniques by leveraging more advanced sensors. Our research group has recently built a prototype car for driverless mobility research, which is instrumented with a plethora of such sensors: cameras, LIDAR, differential GPS, driving control signals, etc. Our aim is to collect driving data in a wide range of realistic scenarios, and to combine it with historical insurance claim datasets recorded by Foyer. Using this data, we will be able to design human-understandable features for safe driving, and to train machine learning models to infer these features from raw sensor measurements. Our risk assessment pipeline will finally be set up on small embedded computers, to be tested and evaluated on board of a fleet of cars from volunteering employees from Foyer Assurances SA.