Optimization-based coordination strategies for connected and autonomous vehicles

Doctoral thesis, 2019

Automated vehicles (AV) are expected to reach the consumer market within the next decade.
Once AVs become ubiquitous, they could resolve difficult traffic situations through communication-based cooperation.
Intersections are of particular interest in this context, as they form bottlenecks in the traffic system and are responsible for a large share of all accidents.
Rather than relying on traffic lights, road signs and rules, AVs could employ cooperative strategies to decide how an intersection should be crossed safely and efficiently.
However, designing efficient coordination strategies for AVs at intersections is challenging, as computationally hard problems are involved, with a safety-critical dependence on both wireless communication and imprecise sensing.

This thesis treats control algorithms for cooperative coordination of CAVs at intersections.
The proposed algorithms are based on Optimal Control (OC) formulations of the coordination problem and aim at finding the optimal control commands for each vehicle through a two-stage approximation procedure.
In the first stage, the order in which the vehicles cross the intersection is determined using a heuristic based on Mixed-Integer Quadratic Programming (MIQP).
In the second stage, the optimal control commands for each vehicle are found under a fixed crossing order.
Two algorithms are presented that solves the problem of the second stage in a communication efficient, distributed fashion.

In the first algorithm, the problem is decomposed into one master-problem and one sub-problem for each vehicle.
The master-problem is solved using a Sequential Quadratic Programming (SQP) algorithm, where most computations are performed in parallel on-board the vehicles.

In the second algorithm, the problem is solved using a Primal-Dual Interior Point (PDIP) method.
The computations involved are separable so that the largest part can be performed in parallel on-board the vehicles, a lesser part in parallel on lead-vehicles for each lane, and a small part at a central network node.

The two-stage approximation procedure is used in a Model Predictive Controller (MPC), and conditions for persistent feasibility and stability are derived.
Performance of the MPC-based closed-loop controller is assessed in simulation, and compared to traffic-lights and alternative coordination algorithms.
The results demonstrate that the two-stage approach outperforms existing alternatives, with almost zero average travel-time delay and a marginal increase in energy consumption compared to cruising at constant speed.

An MPC controller based on the SQP algorithm is verified experimentally at a test-track with three real vehicles.
The results demonstrate that efficient coordination is practically realizable through communication-based optimization and MPC.
In particular, the experiments show that the MPC algorithm performs well under adverse conditions with significant sensor noise, communication impairments and external perturbations.