Current AV techniques do not scale. To truly bring AVs to all parts of the globe we need unconventional methods. In this talk I will present novel approaches for scalable AVs. First I will discuss how we autolabeled the nuPlan dataset for ML-based planning. Then I discuss a warm start technique called BASAL for active learning for lidar segmentation. Finally I will discuss the HORUS dataset that enables us to study drone-vehicle communication for improved safety.

The recent breakthroughs in using deep neural networks for a large variety of machine learning applications have been strongly influenced by the availability of high performance computing platforms. As a result, today's pre-production self-driving vehicles require high-performance central computing platforms with an electrical power consumption of more than 4000 W. There is an urgent need for a holistic AI system hardware/software codesign approach that enables optimized deployment of multiple machine learning workloads on AI hardware accelerators with minimized energy and resource demand. This talk will discuss techniques for optimized deployment and hardware/software codesign for commercial off-the-shelf and custom accelerator architectures.

This talk will explore recent advances in video-based driver observation techniques aimed at creating adaptable, data-efficient, and uncertainty-aware solutions for in-vehicle monitoring. Topics covered will include: (1) an overview state-of-the-art  methods and  public datasets for driver activity analysis  (2) the importance of adaptability in driver observation systems to cater to new situations (environments, vehicle types, driver behaviours) as well as strategies for addressing such open world tasks, and (3) incorporating uncertainty-aware approaches, vital for robust and safe decision-making. The talk will conclude with a discussion of future research directions and the potential applications of this technology, such as improving driver safety and improving the overall driving experience.

As robotaxi enters the commercialization phase, one of our main areas of focus is perception generalization, with respect to novel moving objects, or changing conditions (e.g., rain) in the wild. In this talk, I will present our pilot explorations along this line of research.