CSE DSI Machine Learning Seminar with Burhan Yaman (Bosch Research)

Autonomous driving is a challenging task that requires both a deep understanding of the environment and the ability to react swiftly to changes. While recent advancements in end- to-end autonomous driving have led to notable improvements in scene understanding, several key challenges remain, including a lack of reasoning, poor generalization performance, and difficulties in handling long-tail scenarios. Meanwhile, recent progress in Foundation Models— such as vision-language models and diffusion models—has demonstrated remarkable common sense reasoning and generalization capabilities. However, integrating these approaches into autonomous driving remains an open challenge due to stringent latency constraints.

In this talk, we will present representation alignment strategies that enable the efficient distillation and integration of foundation models into autonomous driving systems without introducing additional computational costs at inference time.