Periodic Reporting for period 1 - COeXISTENCE (Playing urban mobility games with intelligent machines. Framework to discover and mitigate human-machine conflicts.)
Reporting period: 2023-03-01 to 2025-08-31
Imagine a city where intelligent machines—like self-driving cars—share the streets with human drivers. Sounds futuristic? It’s not far off. But as machines become smarter and more widespread, we face a major question: how will humans and machines coexist in complex environments like urban traffic?
The COeXISTENCE project tackles this very challenge. Its goal is to understand and shape the interaction between humans and intelligent machines when both compete for limited urban mobility resources, such as road space. The project focuses specifically on the way vehicles choose their routes in a city and explores what happens when autonomous vehicles (AVs) use artificial intelligence (specifically, multi-agent reinforcement learning) to learn optimal strategies.
To do this, the team outlined four strategic steps: simulate, discover, assess, and mitigate. These range from building advanced simulations of traffic mixing AVs and human drivers, to identifying social risks and possible unfairness that may arise when intelligent machines act selfishly. The project goes beyond simply modeling traffic—it delves into equity, social dynamics, and how to ensure fair and efficient systems.
Ultimately, COeXISTENCE doesn’t just aim to answer how AVs should choose their routes. It wants to define a whole new research area—where mobility, artificial intelligence, and social fairness intersect—and to offer practical tools and theoretical insights for a more harmonious human-machine future on our roads.
The COeXISTENCE project tackles this very challenge. Its goal is to understand and shape the interaction between humans and intelligent machines when both compete for limited urban mobility resources, such as road space. The project focuses specifically on the way vehicles choose their routes in a city and explores what happens when autonomous vehicles (AVs) use artificial intelligence (specifically, multi-agent reinforcement learning) to learn optimal strategies.
To do this, the team outlined four strategic steps: simulate, discover, assess, and mitigate. These range from building advanced simulations of traffic mixing AVs and human drivers, to identifying social risks and possible unfairness that may arise when intelligent machines act selfishly. The project goes beyond simply modeling traffic—it delves into equity, social dynamics, and how to ensure fair and efficient systems.
Ultimately, COeXISTENCE doesn’t just aim to answer how AVs should choose their routes. It wants to define a whole new research area—where mobility, artificial intelligence, and social fairness intersect—and to offer practical tools and theoretical insights for a more harmonious human-machine future on our roads.
Over its first phase, COeXISTENCE has laid the groundwork for a new generation of mobility research. The team developed RouteRL, a powerful and user-friendly framework that lets researchers simulate cities where humans and autonomous vehicles coexist. It’s not just a toy model—it includes real traffic dynamics, customizable agent behavior, and is delivered plug-and-play Python package used by the broader scientific community.
But the achievements didn’t stop at simulation. In controlled experiments, researchers showed how AVs using selfish strategies can unintentionally harm human drivers. For instance, if just a few AVs act purely to minimize their own travel time, they can make traffic worse for everyone else—unless carefully coordinated.
The team also explored how today’s AI algorithms can struggle in complex environments. In a position paper, they argued that when multiple AVs learn simultaneously, it can lead to unstable traffic systems unless properly managed. It’s not that the AI isn’t smart—it’s that the environment becomes too unpredictable, especially when human behavior is added to the mix.
For the future governance of mixed human-Av systems, we propose a novel concept of Wardropian Cycles—traffic assignments that can be both fair and efficient over time. By rotating which routes drivers use each day, it’s possible to achieve both optimal traffic flow and social equity, something long thought to be impossible. With AVs offering the potential for precise control, this concept might soon become reality.
But the achievements didn’t stop at simulation. In controlled experiments, researchers showed how AVs using selfish strategies can unintentionally harm human drivers. For instance, if just a few AVs act purely to minimize their own travel time, they can make traffic worse for everyone else—unless carefully coordinated.
The team also explored how today’s AI algorithms can struggle in complex environments. In a position paper, they argued that when multiple AVs learn simultaneously, it can lead to unstable traffic systems unless properly managed. It’s not that the AI isn’t smart—it’s that the environment becomes too unpredictable, especially when human behavior is added to the mix.
For the future governance of mixed human-Av systems, we propose a novel concept of Wardropian Cycles—traffic assignments that can be both fair and efficient over time. By rotating which routes drivers use each day, it’s possible to achieve both optimal traffic flow and social equity, something long thought to be impossible. With AVs offering the potential for precise control, this concept might soon become reality.
We hope the breakthrough is still in front of us. So far we identified and reported novel and significant issues, created a software to experiment with this and propose solutions. Which is presumably beyond-state-of-the-art, yet for a breakthrough in future we hope to leverage on this and provide a seminal contribution, either in the field of algorithms, findings or methods.