Examenspresentation av Andreas Gelotte

Examinator:
Jonas Sjöberg

Supervisors:
Hasith Karunasekera, Chalmers
Magnus Gustafsson, CEVT

Company:
China Euro Vehicle Technology AB (CEVT)

Abstract:
An approach to estimating road friction and corresponding passenger vehicle braking and cornering performance is presented. In particular, it classifies dry and wet asphalt, gravel, snow- and ice-covered roads using an onboard front-facing camera. To estimate road surface condition, the convolutional neural network SqueezeNet is retrained using 4100 images of the mentioned road conditions. All five classes are associated with individual tyre slip-friction relations as per Pacejka's Magic Formula. A final estimation of surface condition is given by an arbiter, taking input from convolutional neural networks, ambient temperature and road surface type databases for the current GPS coordinates. Models for minimum braking distance and safe cornering speed are proposed using point mass modelling for the current road friction estimate.

The model is verified in vehicle on automotive proving grounds and public roads, evaluating road surface condition classification performance in steady-state driving and road friction estimation during dynamic manoeuvres. Predicted braking distances are within 22% of experimental data, and safe cornering speeds within 3%. Surface type is correctly identified in most live vehicle tests, while classification performance reach 78% for pre-recorded images of roads."

Welcome!

Andreas, Hasith, Magnus and Jonas