Rudolf Reiter

University of Freiburg

Tuesday, November 12, 2024, 14:00

SR 02-016/18, Geb. 101

Abstract

This thesis addresses the challenges of motion planning and obstacle avoidance in autonomous driving, particularly focusing on the complexities introduced by nonconvexity, high-dimensional planning spaces, real-time requirements, and the unpredictable behavior of other vehicles. It proposes optimization-based methods to tackle these issues effectively, using nonlinear and combinatorial techniques enhanced by machine learning. The research concentrates on three main areas.

Vehicle Modeling for On-Road Driving: It introduces an optimized vehicle model using road-aligned coordinates, improving computational efficiency through sequential quadratic programming and a novel "lifted" formulation for obstacle avoidance in transformed and Cartesian states.

Nonconvex Obstacle Avoidance: To address the inherent nonconvex nature of obstacle avoidance, the thesis uses mixed-integer optimization, which combines discrete and continuous planning. This approach mitigates computational complexity by adapting the problem to various settings like static environments and structured highway driving and uses machine learning to predict discrete assignments for real-time applicability.

Interactive Planning in Competitive Scenarios: For competitive driving scenarios, such as in autonomous racing, the thesis presents real-time algorithms combining reinforcement learning and model predictive control to predict and influence the behavior of other vehicles.

The thesis advances autonomous driving planning by developing models that balance computational efficiency, safety, and adaptability in complex environments.

Committee

Vorsitz:    Prof. Dr. Stefan Rupitsch
Beisitz:    Prof. Dr. Joschka Bödecker
Prüfer:    Prof. Dr.  Moritz Diehl (Betreuer / Erstgutachter)
Prüfer:    Prof. Dr.  Melanie Zeilinger (Zweitgutachter)