The development of optimization-based control strategies for autonomous distributed multi-agent systems is at the core of the project. A specific example for such a task is road traffic scenarios, where autonomous cars communicate with each other and "negotiate" control actions that allow to accomplish the individual tasks of each car but avoid collisions at all time. The aim is to develop a control framework that is capable of controlling a distributed system of cars automatically while taking into account constraints and individual objectives. The interactivity, the dynamics, and the complex environment render this task a challenge and several issues have to be addressed on the algorithmic level and with regard to the implementation on real autonomous (scale) cars. We address
- modeling issues (environment, roads, car models, driver model by optimal control),
- tracking controller design,
- coordination of agents in interconnected scenarios,
- implementation of control algorithms,
- collision avoidance through state constraints.
Our concept is based on the repeated solution of suitably defined optimal control problems or generalized Nash equilibrium problem in a nonlinear model-predictive control framework. The optimal control problems are being solved by our direct shooting method OCPID-DAE1 or by dynamic programming, if the dimensions are sufficiently small.
Current research topics:
- coordination of multiple vehicles in road networks
- generalized Nash equilibria as a model for cooperation
- hierarchical distributed model-predictive control