We are currently doing research on Autonomous Cars and Simulations of V2X-Communications. To evaluate several different approaches in case of intersection management, we are looking for hints and new ideas.
Fully decentralized means you don't have any global signal for coordination. A (usual) technique is to use swarm algorithms. This works well at low vehicle entrance rates (i.e. flows are small). It becomes much more tricky at higher levels, when congestion becomes a problem.
In previous papers we have proved that you need at least to reach an agreement for each pair of interacting vehicles (what we call priorities) : then you can have a distributed control (integrators, MPC or other). This means precisely you have 2 different controls for the 2 priorities (before or after the other vehicle). If you want to have a unique control (and this control to be continuous), we can prove some initial states lead to deadlock (and inefficiencies if you are close to deadlock) : my last ITSC paper (about decision and control).
Finally this lead to the difficult question to first set priorities at a global scale (we call it a priority graph). And then to find a distributed control for the cars. We have shown a way to solve efficiently the second problem (MPC relying on the priority graph). The first problem has 2 sides : what is an efficient priority graph? how can we build it in a distributed way. Again there are solutions but it is ongoing research.
If you are interested, I can point you the relevant papers.
you might visit the website of the German government-funded project "UR:BAN". Several approaches on different scales (regional networks, urban structures and the smart intersection itself) were covered. Different technologies were considered as well, inter alia V2X based on Wifi 802.11p.
Thank you very much for your responses Ulrich and Arnaud! I will have a look at them and keep in mind to contact you in case of further questions. I really appreciate your help.
Thank you very much for your input! I Uploaded the results of our multiple AIM approaches on my profile, feel free to read it, I am always open for discussions.