Autonomous Vehicle Decision Making and Game Theory
Each passing year, fully autonomous vehicles seem more and more likely to enter the public market, with manufacturers such as Tesla and Waymo consistently making technological leaps, removing previous obstacles. While there is natural and understandable reason for concern for the everyday consumer regarding self-driving cars – safety and privacy being the most significant – self-driving cars are also able to bring many benefits to every driver. One such benefit is the reduction of a common pest for any driver: traffic.
Researchers from the University of Tuzla and the University of Louisville have utilized a game theory approach to develop a vehicle-to-vehicle decision making algorithm to decrease traffic congestion at roundabout intersections. Roundabout intersections, while not yet common in the United States, are already popular in many countries around the world and are becoming increasingly popular in the US, as they have been shown to be both safer and faster than traditional stop-sign or traffic light controlled intersections. However, if traffic is heavy, waiting time at a roundabout can become quite large, thus severely slowing down the efficiency of the roundabout. Autonomous vehicles can greatly lessen this inefficiency by reducing the gaps between cars, which is what the reseachers were seeking to do.
They first modelled the problem as a Prisoner’s Dilemma with two players – one car already circulating the roundabout, and one car attempting to enter. By considering the options for the circulating vehicle as “smoothly exclusion” and “smoothly circulate”, and the options for the entering vehicle as “smoothly inclusion” and “adjusting speed”, the researchers were able to create a payoff matrix for the two players, and by a simple geometric and physical argument, they were able to deduce the payoffs for each option. By analyzing this simple payoff matrix, the researchers were easily able to make an algorithm to efficiently manage roundabout logic for two players. Using two robots, they then experimentally verified their algorithm by testing it in a variety of different scenarios.
There is still much to be desired from the results of this study, which the researchers themselves acknowledge – specifically that in real life traffic scenarios, “additional factors may shape
the cooperate model between more vehicles”. Nonetheless, the study is an important milestone in the development of autonomous vehicle logic, as it demonstrates that seemingly complex problems can be elegantly modelled by applying game theory principles.
Sources:
https://pdfs.semanticscholar.org/b17e/d95d51e8669b8f7e37af2519a8df796fd6b5.pdf
https://wsdot.wa.gov/Safety/roundabouts/benefits.htm