A0367
Title: Improving motion prediction in autonomous driving with expert knowledge: a Bayesian deep learning approach
Authors: Christian Schlauch - Co-Pace GmbH (Germany) [presenting]
Abstract: Autonomous driving is one of the most highly anticipated yet elusive mobility innovations. The field has made significant advances through deep learning, especially in perception and motion prediction. Still, the field faces open challenges since safety requirements in autonomous driving demand robust domain adaptations between locations and well-calibrated uncertainty estimates for the numerous high-risk edge cases in urban environments. To meet those demands, the potential of Bayesian deep learning methods is explored for motion prediction. It demonstrates how a Bayesian approach can be used to regularize commonly employed motion prediction models by utilizing prior expert knowledge. More specifically, a CoverNet baseline model is adopted with a compute-efficient last-layer Gaussian Process approximation, and prior drivability knowledge is integrated. Doing so improves both robustness and calibration, as evaluated on various datasets.
