Full Throttle Vision: Vehicle detection for autonomous racing

Institut

Lehrstuhl für Fahrzeugtechnik

Typ

Semesterarbeit / Masterarbeit /

Inhalt

experimentell /  

Beschreibung

Autonomous driving is a fast-moving research discipline. At the moment, most of the research community’s attention is centered on public streets. However autonomous racing provides a few benefits for research: High-speed driving uncovers a different set of problems than the usual public road scenes. At the same time, the closed track allows for testing new algorithms without safety concerns for the public. The Chair of Automotive Technology competes in the Indy Autonomous Challenge where autonomous Dallara vehicles race against each other at speeds exceeding 250 km/h. At these hight speeds, perception algorithms need to detect other vehicles faster and more accurately to precisely determine other vehicles movements. The lower the delay from sensor to detection, the less compensation needs to be done for the delay and more accurate measurements can be used in the rest of the software stack. However, current deep-learning approaches don’t focus on latency primarly, which is why they can be relatively slow (>100ms). The goal of this thesis is to develop accurate, low-latency detection pipelines for autonomous racing.

• Research into low-latency perception algorithms
• Development of a low-latency perception pipeline
• Benchmarking against current approach used by the chair and state-of-the-art
• Focus on Camera-based pipelines to add redundancy to current LiDAR approach

Voraussetzungen

Requirements for this thesis are:

• Experience with Python and especially PyTorch
• Highly motivated to experiment and try different things
• Some experience with Deep Learning Algorithms
• Knowledge about autonomous driving preferable
• Experience with ROS 2 and Docker are a bonus

Tags

FTM Studienarbeit, FTM AV, FTM Ebner

Möglicher Beginn

sofort

Kontakt

Dominic Ebner, M.Sc.
Raum: MW3510
Tel.: +49.89.289.15871
dominic.ebnertum.de