Automation of road intersections using distributed model predictive control

The automation of road intersections is increasingly considered as an inevitable next step toward a higher level of autonomy on our roads. For the particular case of fully automated vehicles, we propose a distributed model predictive control approach in which multiple agents are able to pass the intersection simultaneously while keeping a sufficient safety distance to conflicting agents. Therefore, each agent solves a local optimization problem subject to non-convex safety constraints which couple the agents. In order to handle these coupling constraints, we propose constraint prioritization. With that methodology, for two pairwise conflicting agents, the safety constraint is only imposed on the agent with lower priority which does not imply any a priori intersection passing order. Finally, we can solve the distributed optimization problem in parallel without any nested iterations. To solve the local non-convex optimization problems, we apply a semidefinite programming relaxation in combination with randomization to obtain appropriate and feasible solutions. A simulation study finally proves the efficacy of our approach.