https://events.vtools.ieee.org/m/556950
Autonomous mobility is largely approached as a vehicle-centric problem. Persistent challenges in safety,
scalability, and public trust suggest a deeper issue: “intelligence” is often considered in isolation rather than as
distributed. This presentation argues that truly safe and trustworthy autonomy will emerge only through
symbiotic computational systems, where perception, decision-making, and control are distributed across
humans, machines, and infrastructure.
The presentation starts with an overview of the four decades-long progress in autonomous driving and related
advancements in driver assistance technologies. It is followed by a discussion of the central thesis: that many
failures in autonomous mobility stem not from algorithms alone, but from how system boundaries are defined—
what is sensed, where intelligence resides, and how responsibility is shared. Framing autonomy as a systems-
level problem, the talk draws on principles of distributed and embodied cognition to unify perspectives from
robotics, artificial intelligence, human–computer interaction, and transportation engineering.
Concrete examples from multidisciplinary research by the CVRR and LISA teams at UC San Diego, conducted
on real vehicles in real-world driving environments and validated through both quantitative benchmarks and
qualitative studies in collaboration with industry partners, illustrate how shared autonomy can tightly couple
human state (e.g., intent, attention, readiness) with environmental context to enable safer and more adaptive
human–AI interaction. The lecture also discusses how advances in foundation models, self-supervised learning,
and active learning can improve generalization and robustness in safety-critical settings.
The talk concludes with key open challenges, including multimodal foundation models for traffic ecosystems,
human–AI co-adaptation, and continual learning under domain shift, important problems to realize scalable,
trustworthy autonomous mobility.
