Highlights from IEEE CDC 2024 — Part 1: Autonomous Vehicles

The IEEE Conference on Decision and Control (CDC) is the biggest annual gathering of control engineers and control theorists from around the world. I enjoyed IEEE CDC 2024 and decided to share the highlights from my perspective, for those who couldn't attend. In this first part I focus on autonomous vehicles, which were a prominent topic this year.

Plenary Talk 1. I suspect most people working on autonomous vehicles would agree that the biggest challenge is the uncertainty in the environment, and how to provide provable safety guarantees given that uncertainty. Two of the four plenaries dealt with autonomous vehicles, and one tackled uncertainty head-on: "Towards Safe and Resilient Autonomy Using Synergistic Control, Observation, and Learning," by Dimitra Panagou (University of Michigan). Her examples focused mainly on drones facing unexpected obstacles on a planned route — for instance a wall placed in their path — and the method her team developed to guarantee safety in such situations, with video clips of the results. She mentioned autonomous vehicles in her introduction, though the examples were drones; when I asked afterward, she said she could share more of her work on that topic, which I look forward to.

Plenary Talk 2. The other plenary was "Transforming Mobility through Learning and Control," by Karl Johansson (KTH), whose work I've followed for years and hold in high regard. He devoted much of it to platoons of cars — vehicles that communicate and coordinate along a route — with examples from his collaboration with Volvo, analyzing the congestion platoons can create on motorways and how to mitigate it. He was enthusiastic and encouraged younger control engineers to pursue the topic. I have to disagree here: except when platoons venture into unknown or hazardous areas, I don't see the benefit if each vehicle can already drive autonomously — and it introduces risk if the inter-vehicle communication is hacked, which is always possible. More compelling was his treatment of a single vehicle approaching a junction where other objects (vehicles, pedestrians) are occluded — incidentally, the best example I can think of for explaining observable vs. unobservable states, since here they are literally observable or not. His solution adds auxiliary devices at each junction that communicate what the vehicle can't see, including a worst-case analysis assuming a vehicle is always occluded behind the visible ones — useful for current industrial applications. The most interesting part, though, was theoretical and experimental work on the microscopic and macroscopic properties of traffic flows, using real-world recordings from routes in Sweden with Volvo and Scania. I've made a note to look into this more closely. His slides are here.

Other notable talks. Regular sessions were also dedicated to autonomous vehicles, of varying quality. Some works were intellectual curiosities rather than of near- or far-future use — a prime example, "A Feasibility Analysis at Signal-Free Intersections," proposed removing traffic lights so a device at each junction takes over and maneuvers all vehicles optimally; I find it impractical for various reasons. Others had real industrial relevance, such as "Uncertainty-Aware Decision-Making and Planning for Autonomous Forced Merging," a motion-planning scheme for a vehicle merging onto a road from a ramp given uncertainty in the estimated acceleration of adjacent vehicles.

There was more. Other talks covered the very important topic of the cybersecurity of self-driving cars, among others. You can search the "Autonomous vehicles" keyword in the conference program to find the abstracts of all the relevant talks.

Part 1 of 3 — originally published on LinkedIn, January 7, 2025.