Boston University Professor Christos Cassandras has won the 2026 IEEE Intelligent Transportation Systems Society (ITSS) Outstanding Research Award for his work on self-driving cars and urban transportation. The university announced on June 1 that the core of his research is a transportation system in which cars and roadside infrastructure share information and coordinate their movements. Following the research described in the university's award announcement leads to a mechanism that lets cars cross intersections without traffic signals.
To cut down on waiting at red lights, what do cars need to share with one another, and where should they slow down? And how far does a mathematically proven "safety" extend to human-driven cars and pedestrians crossing the road? The related papers reveal both how such a system reduces stops and the conditions under which its safety guarantees hold.
Adjusting the time of entry into the intersection, well before arrival
Cassandras's laboratory has simulated cars crossing without signals, using the intersection of the Boston University Bridge and Commonwealth Avenue as its model. A 2016 university article includes a video in which cars adjust their acceleration and deceleration ahead of the intersection and pass through without coming to a complete stop. This award recognizes that long-running body of work.
A conventional traffic signal divides up time on the intersection by alternately giving crossing routes a green light. If cars can communicate with one another, that allocation of time can be adjusted much more finely.
For example, cars arriving from one direction are let through first, while cars from another direction slow slightly beforehand and enter the intersection afterward. Rather than driving fast right up to the intersection and then stopping, cars adjust their speed while moving so that their arrival times fit together.
A 2016 preprint by Yue Zhang, Cassandras, and Andreas Malikopoulos formulates this coordination mathematically for adjacent intersections.
In a "control zone" set up ahead of the intersection, each car decides its acceleration based on its own position and speed. It keeps a safe distance from the car ahead in the same lane, and cars whose paths cross coordinate so that the times they occupy the intersection do not overlap.
In other words, the conditions for preventing rear-end collisions and for preventing lateral collisions are handled separately.
The laboratory's description calls this approach decentralized optimal control. Cars share the necessary information with one another, and each car computes its own acceleration and deceleration.
In a 2018 peer-reviewed paper, the roadside coordinator only assists with passing information along and is not involved in the control decisions of individual cars.
If the number of stops and restarts can be reduced, large changes in speed can be reduced as well.
However, to arrive at the intersection at the assigned time while also keeping safe distances along the way, the speed and timing at which a car enters the control zone must be appropriate.
The 2016 preprint also proposes setting up an additional zone further upstream of the control zone to establish those conditions. Preparation for safe passage through the intersection begins before cars ever enter it.
To prove safety, the preconditions must be met
A 2018 paper in Automatica by Malikopoulos and colleagues derives conditions under which a solution satisfying the safety constraints exists, and presents a method for determining cars' acceleration and deceleration.
The basic problem setup, however, rests on several assumptions.
All cars are self-driving vehicles with identical characteristics that can coordinate via communication, and right turns, left turns, and lane changes are not considered. The paper also assumes that cars maintain a constant speed within the intersection and that they can obtain the necessary local information without error or delay.
Preventing rear-end collisions inside the intersection requires further conditions.
In its conclusion, the paper lists as future work rear-end collisions that could occur when cars in the same lane enter the conflict zone at different speeds.
The authors suggest as candidate approaches solving the optimal control problem after aligning entry speeds at the intersection, or relaxing the assumption of constant speed inside the intersection and introducing control that prevents rear-end collisions.
Having each car travel at a constant speed does not, in itself, eliminate the speed difference between the cars in front and behind.
Extensions to cases with communication or measurement errors and delays are also under consideration.
The authors state that if upper bounds on the size of errors and delays are known, the idealized assumptions can be relaxed by setting the safety constraints more conservatively.
Conversely, determining how far safety can be guaranteed requires quantifying how much uncertainty exists in the sensors and communications.
Being able to find a safe solution mathematically and real cars being able to satisfy the underlying assumptions are separate matters that must be verified separately.
The paper also compares the approach with fixed-time signal control in a simulation modeled on a Boston intersection, covering 448 cars.
Each direction has one lane, and cars arrive as a Poisson process at 450 vehicles per hour per lane. The control zone is set at 245 m, and the zone where car paths cross is set at 35 m.
Using the traffic simulator VISSIM and estimating fuel consumption from speed and acceleration, the paper reports improvements of 46.6% in fuel consumption and 30.9% in travel time compared with fixed-time signal control.
However, this is a simulation comparison under conditions in which all cars move cooperatively.
Interactions between two intersections are also not included in this verification.
The paper presents the comparison results in a figure, but the text does not list absolute before-and-after values for fuel quantity or travel time.
Therefore, these reduction rates cannot be taken out of context and interpreted as meaning the same effects would be obtained on ordinary roads today.
The mathematical proof guarantees safety only within the traffic model and safety constraints that were set.
Extending those assumptions to real roads raises the question of how to handle human-driven cars and pedestrians.
What changes when human-driven cars and pedestrians join in?
A 2025 paper in Automatica by Anni Li, Andres S. Chavez Armijos, and Cassandras addresses lane changes in mixed traffic of self-driving and human-driven cars.
A manuscript made public in 2024 assumes two cooperating self-driving cars and one human-driven car.
When a self-driving car moves in front of a human-driven car, it adjusts its trajectory while estimating how that driver will react.
When it moves in front of a self-driving car that can cooperate, on the other hand, matching speeds with that car reduces reliance on predicting human reactions.
The idea is not to predict human decision-making perfectly, but to make use of the relationship between self-driving cars that can coordinate directly.
For the safety constraints, the researchers use "control barrier functions."
An elliptical safety region whose size changes with speed is set around each car, and constraints are placed on acceleration and steering so that distances to other cars are maintained.
When a planned trajectory would violate those constraints, the control input is corrected.
However, an upper bound is also set on the error in estimating the state of the human-driven car. The guarantee does not hold regardless of whatever sudden maneuvers a human might make.
At intersections with pedestrians, there are even more things to control.
A 2023 preprint by Yingqing Chen and Cassandras proposes a method of adjusting signals that handles the waiting times of vehicles and pedestrians together.
Even if the flow of cars alone is efficient, it is hard to say the road is being shared fairly if pedestrians are kept waiting for long periods.
This control sets upper and lower limits on green-light durations, and also sets an upper limit on pedestrian waiting time.
The abstract of a peer-reviewed paper by the same authors, published February 2, 2026, likewise states the goal of signal control that lets vehicles and pedestrians share an intersection fairly.
It uses actual traffic data obtained in Veberöd, Sweden, but the evaluation itself was carried out on a simulated intersection.
Even research that keeps signals leads to the same challenge: how to coordinate the movements of the various users who take part in traffic.
Control of signal-free intersections, lane changes in mixed traffic with humans, and signal control including pedestrians are studies that differ in both their subjects and what they control.
Organizing Cassandras and colleagues' papers by subject, controlled quantity, and verification method gives the following.
| Study | Main subject | Quantity controlled | Verification and scope |
|---|---|---|---|
| Signal-free intersection, 2018 | Self-driving cars that all coordinate via communication | Intersection entry time and acceleration/deceleration | Theory and traffic simulation. The basic setup assumes straight-ahead travel; right and left turns and lane changes are excluded |
| Lane changes in mixed traffic, 2024 preprint / 2025 publication | Two self-driving cars and one human-driven car | Lane-change trajectory and control inputs that maintain the safety region | Theory and simulation. An upper bound is set on the error in estimating the human-driven car's state |
| Signal control for vehicles and pedestrians, 2023 preprint / 2026 publication | Traffic flows of vehicles and pedestrians | Green-light durations and criteria related to queues and waiting times | Evaluated at a simulated intersection. For the 2026 paper, the published abstract was consulted; details are based on the 2023 preprint |
This comparison classifies the problem setup and evaluation method of each paper.
It is based on Sections 2 and 4 of the 2018 paper, Sections 2 and 7 of the mixed-traffic preprint, Sections 2 and 4 of the pedestrian preprint, and the abstract of the 2026 paper.
Because the road geometries and the traffic participants differ, performance cannot be simply compared side by side.
Even if safe movement is demonstrated for lane changes in mixed traffic, that alone does not prove the safety of a signal-free intersection that includes pedestrians.
Whether communication between cars can be trusted is also a safety condition
In systems where cars travel while sharing their positions and speeds, how far the information they receive can be trusted bears directly on safety.
A 2024 preprint on communication security by H. M. Sabbir Ahmad and colleagues addresses cooperating vehicles at intersections and similar settings, covering attacks that register nonexistent vehicles and attacks that inject hard-to-detect false information.
It proposes a method that updates a trust level for each vehicle and combines it with control barrier functions that account for errors.
This research also has preconditions, however.
It assumes that the roadside coordinator itself is not attacked, places an upper limit on the number of fake IDs an attacker can create, and excludes attacks in which the attacker deliberately crashes its own vehicle.
Safety guarantees that include countermeasures against attacks likewise hold only within the assumed types of attacks and information conditions.
A system that keeps traffic flowing without stopping requires not only vehicle control but also an environment in which the shared information can be trusted to a certain degree.
Getting cars to cooperate also requires incentives
Some research also provides human drivers with reasons to cooperate.
The university's CISE announcement of the award describes a research concept in which cars that yield the road to improve overall traffic are given small digital rewards.
In a system where only certain drivers bear the burden of yielding while others reap the benefits, it is hard to sustain cooperation over the long term.
Both enabling cars to communicate with one another and creating mechanisms that make road users want to cooperate are design challenges for a transportation system.
To judge whether traffic lights can be eliminated, we need to confirm that safe spacing can be maintained even on roads where some cars and pedestrians do not cooperate with the system.
We must also evaluate how the system behaves when communication errors or delays occur, and whether particular users are made to wait for long periods.
If these conditions can be verified in the same road environment, coordination among cars may move beyond a technology for simply moving platoons faster and toward traffic control that reduces stops and waiting times for a wide range of road users.