Smart Car Researcher Wants to Eliminate Stoplights

We have all spent way too much of our finite lives sitting at red lights in empty intersections. It is one of those small, daily injustices that points to a larger problem: our infrastructure is still running on 20th-century logic. We have computers in our pockets and advanced sensors in our bumpers, yet we still rely on colored lightbulbs to tell us when it is safe to move.

Christos Cassandras, a professor at Boston University and a heavy hitter in systems engineering, has spent four decades looking at this exact friction. He is not just complaining about the commute; he is building the mathematical framework to kill the stoplight entirely. For builders in the crypto and AI space, his work is a masterclass in how we transition from "dumb" centralized systems to decentralized coordination.

The Shift from Physics to Events

To understand where Cassandras is going, you have to understand where he started. In the 1980s, while working on a project for Fiat, he realized that the way we modeled the world was broken for machines. Most engineering at the time relied on classical physics—continuous time, smooth motions, the kind of math Newton would recognize. But a car assembly line, or a computer, does not work like a rolling ball. It works in "events."

A machine sits idle until a part arrives. A computer waits until a button is clicked. Cassandras pioneered what we now call Discrete Event Systems. He later merged this with physical math to create Cyber-Physical Systems (CPS). This is the foundation of how we talk about modern hardware today: it is an object that obeys gravity and momentum, but it is also a node that obeys logic and data inputs.

For those of us building in the digital world, this sounds obvious. But in the physical world of civil engineering, this was a radical departure. It moved the focus from "how do we build a bigger road?" to "how do we optimize the logic of the objects on the road?"

Why Stoplights Are a Failure of Coordination

Stoplights exist because humans are unreliable actors. We cannot be trusted to negotiate an intersection at 40 miles per hour with three other drivers simultaneously. The light is a crude, centralized throttle designed to prevent us from killing each other. It works, but it is incredibly inefficient.

Cassandras’s recent work focuses on removing the throttle. Using mathematical models that account for a vehicle’s position, mass, and velocity, he has simulated a world where cars communicate and coordinate their arrival at an intersection. In his simulations of notorious bottlenecks in Boston, removing the lights and letting the cars "negotiate" the space led to smoother flow, higher safety, and lower fuel consumption.

The skepticism here is obvious: what happens when the computer glitches? Cassandras is tackling this through what he calls "Control Barrier Functions." Essentially, he is trying to prove safety mathematically before a car ever hits the pavement. He wants to move away from the "move fast and break things" ethos of early self-driving tech and toward a "prove it works, then build it" founder mindset.

The Hybrid Reality

As a founder, I am always wary of "all-or-nothing" tech visions. We are not going to wake up tomorrow and have every car on the road be fully autonomous. We are entering a messy, multi-decade transition period where high-tech Teslas share the road with 1998 Honda Civics. This is where the real engineering challenge lies.

Cassandras is not just dreaming of a 100% autonomous future. He is looking at how even a small percentage of connected vehicles can act as "anchors" to improve traffic flow for everyone. It is similar to how a few honest nodes in a network can improve security for the whole system. You do not need total adoption to start seeing the benefits of decentralized coordination.

"Ultimately, the key word is safety. You would not buy a self-driving car unless the manufacturer could guarantee that it’s safe for you."

This quote from Cassandras hits on the core barrier for all emerging tech, whether it is AI agents or autonomous cars: trust. You can have the most elegant algorithm in the world, but if the user feels like they are gambling with their life (or their data), they will stick with the legacy system every time.

The Founder’s Takeaway

There are three big lessons for builders here. First, look for the "red lights" in your industry. Where are we using outdated, centralized throttles just because we have not built a better coordination layer? In finance, it is the T+2 settlement cycles. In AI, it is the massive centralized silos of data.

Second, solve for the edge cases before the happy path. Cassandras spent his career on the boring, difficult math of safety because that is the only thing that scales. If you are building AI agents, do not just show me what they can do when everything goes right; show me how they fail safely when the network goes down.

Finally, remember that infrastructure is a social contract. Cassandras’s work with the city of Boston taught him that tech does not exist in a vacuum. If an autonomous intersection makes traffic better for the wealthy but ignores the pedestrians or those in older cars, it is a failed product. We are building tools for humanity, not just for the sake of the technology itself.

The stoplight is a hundred-year-old solution to a human coordination problem. We are finally at the point where we can replace that hardware with software. It will not happen overnight, but the math is already there.

Read the original at IEEE Spectrum →