Reinforcement Learning

Our motivation

Congested streets, noise, exhaust fumes... This is a well-known problem in many inner cities. Inner-city traffic is mainly controlled by traffic signals or traffic lights. The control procedures for these traffic lights are based on traffic engineering rules. However, since traffic is dynamic and stochastic, these manually coded rules cannot control traffic optimally in all situations. 

Can artificial intelligence provide a remedy here? 

This is exactly what we are investigating with other partners in the research project

„Künstliche Intelligenz für Lichtsignalanlagen“ (KI4LSA).

Our idea

The technology we use for this is reinforcement learning. This is a subarea of machine learning in which a software agent (e.g., a traffic light controller) interacts with its environment (e.g., traffic at the intersection) and thereby achieves an optimization goal (e.g., reduction of waiting times) in a self-learning manner. Considerable breakthroughs have been made in this area in recent years.

For example, with the help of reinforcement learning, for the first time ever a computer was able to beat the world champion in the extremely complex board game GO. 

As a result of such successes, reinforcement learning is also increasingly finding its way into industry. The VDMA has already published a guideon the introduction strategy of reinforcement learning in industrial practice in 2019.

Your benefit

The application possibilities are manifold: factory automation, production planning, robotics and smart grid to name just a few. If you would like to implement reinforcement learning projects, Fraunhofer IOSB-INA will be happy to assist you as a competent partner.

Partner

• Stührenberg GmbH, Detmold
• Cichon Automatisierungstechnik GmbH, Pfinztal
• Stadtwerke Lemgo GmbH, Lemgo
• Fraunhofer IOSB-INA, Lemgo