Multi-agent control of a fleet of cybercars
|Project members:||Dr. R. Luo, MSc (Renshi), dr.ir. T.J.J. van den Boom (Ton), prof.dr.ir. B. De Schutter (Bart)|
|Keywords:||Distributed and large-scale systems, Multi-level and multi-agent control|
Cybercars are road vehicles with fully automated driving capabilities. A fleet of such vehicles forms a managed transportation system for passengers, with on-demand and door-to-door capabilities. Therefore, personal mobility will be greatly enhanced if cybercars were widely used in the future.
Although automated driving technology has been developed for individual vehicles, the lack of efficient control strategies for a fleet of vehicles is still one of the biggest challenges that cybercars are facing. Actually, centralized control is tractable when the number of vehicles is small. However, for reasons of scalability, fast computation, and robustness, a centralized control strategy will not be tractable for large-scale deployments. In order to overcome this challenge, a distributed control method where cybercars cooperate with each other is needed.
When it comes to cooperation, cybercars can be characterized as decision-making agents with extensive on-board processing capabilities and communication as well as abundant information of the environment. Our aim is to develop efficient distributed control method for cybercars where on-board controllers optimize the routes of all vehicles in a distributed and cooperative way and the total time spent (TTS) of passengers and total energy consumption is minimal.
Model predictive control (MPC) is widely recognized as a high performance control approach that can be used to determine optimal control actions for constrained systems. It determines these control actions by solving a constrained finite-horizon optimal control problem in a receding horizon fashion. In our research, MPC is used as base to develop efficient distributed control methods.
The route that we follow to conduct our research is:
1. Describing the overall problem using hybrid model.
2. Decomposing the overall problem such that the subproblems are tractable and the resulting solution is as close as to the globally optimal solution.
3. Solving the resulting optimization problem in a distributed way.
4. Developing new approaches for distributed MPC.