UC Berkeley

Contract-based design is a method to synthesize distributed control strategies for large-scale networks of dynamically coupled systems. We propose a framework for using dynamic contracts for controlling traffic networks from temporal logic specifications. Given a traffic network, we partition it into a number of smaller sub-networks and compute a collection of assume-guarantee contracts for their dynamical interconnections. A central supervisor chooses the best contracts optimally according to the current state of the system. We use model predictive control (MPC) to find control sequences optimally for each sub-network subject to its contract obligations to and promises from its neighboring sub-networks. Our method is correct-by-design. A case study on a mixed urban-freeway network is presented, where the objective is infinite-Time congestion avoidance and temporal requirements on the traffic lights in the urban intersections.