Model-based ego-motion and vehicle parameter estimation using visual odometry

Ego-motion estimation based on images from a stereo camera has become a common function for autonomous mobile systems and is gaining increasing importance in the automotive sector. Unlike general robotic platforms, vehicles have a suspension adding degrees of freedom and thus complexity to their dynamics model. Some parameters of the model, such as the vehicle mass, are non-static as they depend on e.g. the specific load conditions and thus need to be estimated online to guarantee a concise and safe autonomous maneuvering of the vehicle. In this paper, a novel visual odometry based approach to simultaneously estimate ego-motion and selected vehicle parameters using a dual Ensemble Kalman Filter and a non-linear single-track model with pitch dynamics is presented. The algorithm has been validated using simulated data and showed a good performance for both the estimation of the ego-motion and of the relevant vehicle parameters.