Generating the Invisible: Generative AI for Scalable Autonomous Driving
Parker Ewen - Torc Robotics
Nov. 28, 2025, 2:30 p.m. - Nov. 28, 2025, 3:30 p.m.
ENGMD 279
Hosted by: Hsiu-Chin Lin
Modern autonomous vehicle (AV) systems increasingly adopt end-to-end architectures, where outputs from perception networks feed directly into downstream prediction and planning modules. This design enables losses from planning objectives to propagate through the entire stack, allowing the system to jointly optimize perception and decision-making. While this approach can yield more efficient and coherent driving policies, it also demands massive, richly annotated datasets—a major bottleneck due to the high cost of data collection and labeling in real-world driving. At Torc Robotics, we are exploring how generative AI can address this challenge by synthetically generating sensor data and annotations at scale. This talk introduces a modern sequential end-to-end AV stack and discusses how generative models can augment traditional data pipelines, accelerating the development of safer and more capable autonomous driving systems.
Parker Ewen is a Research Scientist on Torc Robotics’ AI Foundations team, working on generative AI and scene representations for autonomous driving. He specializes in sparse inductive biases to build efficient models and accelerate dataset augmentation for end‑to‑end learning. Parker has published in leading robotics and mathematics venues including ICRA, IROS, RA-L, SIAM, RSS, and TR-O.
