Project KISLEK

Problem definition 

A key lever for improving traffic flow in inner-city areas lies in optimizing the control of traffic light systems (LSA). In a previous project (KI4LSA), a traffic light control system based on reinforcement learning (RL) AI technology was successfully used in the field for the first time worldwide. The AI solution reduced travel times by around 10%. In KISLEK, the optimization potential for several interconnected traffic junctions is now to be evaluated using RL-based control.

Project goal

The aim of the project is to develop an RL-based traffic light control system for connected traffic junctions that optimizes traffic flow across all junctions. Several traffic lights that are spatially connected influence each other, so that a holistic view is expected to have greater optimization potential than a view of the individual nodes. The solution is to be evaluated in a realistic simulation model of a busy street in the city center of Bremerhaven with real traffic data and compared with a conventional traffic light control system.

Implementation

The selected route section is mapped in a realistic simulation model. Furthermore, an RL algorithm is developed to control the interconnected nodes and an interface to the simulation model is implemented. A safety controller to be integrated into the simulation model interacts with the algorithm and guarantees compliance with all safety-relevant requirements. Finally, the self-learning algorithm is trained in the simulation and then evaluated.

Profile