AASS Seminar - Advancing deep learning-based driver and road user intent recognition

The research centre AASS arranges a seminar with Koen Vellenga, Volvo Cars and University of Skövde.

Remote attendance: https://oru-se.zoom.us/j/65096623293

Abstract

In this presentation I will introduce our latest research on intention recognition systems for advanced driver assistance systems and autonomous driving systems. In our research we explored the following questions: How well do ego-vehicle driver intention recognition systems generalize across geographical locations and diverse sensor setups? Does training on sensors that are unavailable during testing improve intention recognition performance? How good are humans at recognizing road user intentions from videos compared to state-of-the-art deep learning approaches? Finally, we empirically evaluated several probabilistic deep learning methods to understand the limitations of our systems better and formalized an approach combining optimization-based pre-training with MCMC sampling.