Master thesis: Graph-based Prediction of Driver Behavior

25.04.2022, Diplomarbeiten, Bachelor- und Masterarbeiten

In this thesis we aim to develop a GNN-based behavior prediction module using information from applying traffic rules. By representing the surrounding environment in a graph-structured fashion, we hope to achieve better generalization abilities than what can be achieved through a hand-engineered feature vector. Intuitively, given how interaction effects between vehicles can be thought of as propagating across the given traffic scene, graphs are well-suited data structures for our problem domain. We use traffic rules formalized in temporal logic as in previous works [3], [4]. The formalization allows one to rigorously evaluate traffic rule compliance of each traffic participant. The thesis builds up on previous works in the area and will partly enhance our implementation of traffic rule evaluation.

Kontakt: eivind.meyer@tum.de