Game theory in connected autonomous traffic – the only solution

June 17, 2021
posted in
written by Marcin Stryszowski

Self-driving cars seem to be the future, but how long will it be until level 5, ‘press-a-button-and-forget’ autonomy becomes the mainstream mode of transport?

Transport research provides a multitude of synchronised, collaborative traffic systems which all promise to deliver 2.5 times larger road throughput. The result? Elimination of congestion, a 20% increase in energy efficiency, clean air and zero accidents, looking very much like this YT video.

Today, it is not a surprise to find a car equipped with a self-drive button. What might be a surprise, however, is that autonomous cars are expected to make congestion even worse. Rather than heading towards efficient traffic, we are about to spend more time in our super-comfy automobiles – should it not be the other way around?

Traffic – the civilised chicken game

Answering this question might get a bit tangled, but nothing formulates complex, multi-agent problems better than game theory. Congestion occurs when streams of cars intersect, which is referred to as a conflict as modelled by the chicken game. Consider two cars heading towards each other. The one that swerves first loses and gets a -1. The winner gets to cruise unimpeded, scoring a 0. However, if nobody chickens out, the cars crash and everyone gets -100.

This is shown in the Matrix below, with Nash Equilibria (NE) marked:

There are two solutions, with either driver chickening out. Changing this means risking a crash, at your fault, so there is no incentive to deviate.

Let’s consider another example. Anyone who has experienced riding a motorbike in Vietnam, the capital of motorized backpackers, knows that the truck always has priority. The truck driver assumes that the biker values their life more than his truck’s bodywork. Let us assume infinity and -5. If a you were on a motorbike heading towards a truck, the solution is obvious: You. Will. Yield. The first observation when a person is driving, they do complex math in their heads on the fly, adjusting their driving tactics to their strategic objective. Imagine trying to drive calmly when you are late for a meeting with your boss.

The second observation is that we need more gentle and humane traffic de-conflicting mechanisms. We do have these in the form of traffic lights. By alternating stop and go signals, they dictate the policy and drivers are glad to oblige, as the risk of crashing is gone. It is a self-enforced de-conflicting mechanism, which whilst very safe is terribly inefficient.

How will autonomous cars interact with human drivers?

This is where the fun starts. The autopilot button in your state-of-the-art vehicle is a meticulously engineered optimal control system that has one job – to drive. Playing games, especially chicken games, is off the table. Partly because our engineering does not take chances when it comes to our health (unlike us) and partly because we have no idea how to address it ethically.

Another issue is communication. Apart from the obvious brake lights and indicators, there is a whole range of signals that drivers use to communicate that automatic cars just miss. From positioning on the lane, flashing lights, sounding the horn, middle fingers, eye contact with pedestrians and verbal outbursts – but this is just speculation. Research does not exist, as engineers trust the black-box AI to figure out how to drive. As a result, Connected Autonomous Vehicles (CAV) just drive steadily on the middle lane, somewhat like a bus. It does not swear and does not respond to flashing or honking, or even police. As the retired Daimler AG CEO noted, CAVs will be bullied by human drivers, unless designed for assertiveness. But who would insure that?

Complexity of the road transport system

So how can we make CAVs genuinely autonomous? Is there a way to create a truly connected, collaborative traffic system? Theoretically, yes. There are over 2 million scientific publications on the subject and someone must have gotten it right.

In practice, someone should do something – but who? You drive a car that you bought from OEM. It is powered by fuel from the petroleum industry so it can drive on government-operated road infrastructures. When interacting with other drivers, everyone adheres to the Traffic Code and crashes are paid for by insurance companies. That is seven stakeholder groups already and we haven’t even mentioned pollution and urban planning yet.

The system is so complex. With all the factors necessary to deliver the public good, it is hard to tell where to start. The vision of hassle-free connected traffic is there, but there is one last question to ask: “who will pay for that?” The research on Public Goods Game is solid – the one who makes the first step towards collaboration gets to pay the bill.

As a result, year by year, we keep spending billions of man-hours in traffic jams, breathe fumes and crash our cars. In the meantime, road operators look for ways to outsource the cost of the connected infrastructure, car manufacturers encrypt their Engine Control Units (ECU) and lawyers say it is impossible to marry CAVs and human drivers. Worst of all, they are all correct.

The road to connected traffic paradigm

However, there is a solution. Chances are you are reading this article on a mobile device, which has countless apps. Some of them even connect with your car. There are insurance apps that track your telemetry and navigation apps that steer you away from congestions. Each with so much valuable information. Yet, each just working on its own.

Imagine one app, that combines all these functionalities. A navigation app, that connects with your engine’s ECU and traffic lights. As you type your destination, you select your time of arrival. A trade-off between time and cost of the journey, like in the picture below, allowing you to fine-tune the parameters of your journey.

Your input to the system, together with your car’s parameters, would allow for a personal journey intention model. Every strategic decision or advice the car makes, would be tailored to your needs and preferences. As you approach traffic lights, you’d know to speed up before red, or to regenerate energy awaiting for green.
Should there be any danger on the road, an oncoming vehicle could connect and share this information ahead.
On unregulated intersections, such an app could connect with the other car and use game theory to negotiate an optimal solution. This app could provide the optimal solution to every manoeuvre.

Internalisation of negative effects of traffic

But what does optimal mean? The best solution for you, might not be the best for others around. Traffic affects everyone. Air pollution, noise, vibration, dust – these are all negative externalities. There are very crude methods of accounting for it, such as the London Congestion Charge which is a flat fee, Stockholm’s entry charge which is higher in rush hours, the Umwelt Zone in Germany, etc. But still, this is fundamentally unfair. Some drive all day in their V8, and some only want to drop their children to school, yet the latter pays twice.

If an app that manages these could communicate with an engine’s ECU, it could even be possible to propose an inherently fair pay-as-you-pollute system to capture noise, pollution, congestion. You’d get to pay just for the trouble you are causing, putting you at ease when on the way to school, and thinking twice before doing night burnouts. In game theory this is called the ‘Wonderful Life utility function’.


This might sound like science fiction, but it’s not. Every single technology or functionality mentioned here exists today, developed by companies you know. The missing ingredient is the incentive to cooperate and integrate, breaking out of the strategic deadlock of the ‘who’s going to pay for this’ game. Thankfully, even here game theory research helps, providing strong evidence for the effectiveness of pre-play communication.


Written based on the findings of my PhD Thesis.