Innovative and efficient

Artificial intelligence (AI) is increasingly permeating our everyday lives – and consuming more and more energy in the process. In view of climate change and resource scarcity, it is becoming increasingly urgent to develop AI applications in a resource-saving and sustainable way. This is a difficult task – especially when large volumes of sensitive data are to be analyzed using modern AI models such as deep neural networks. These models are not only very complex, they are also used very frequently. This results in high energy consumption and a large ecological footprint. The L3S project Green AutoML for driver assistance systems (GreenAutoML4FAS) aims not only to exploit the advantages of modern AI, but also to optimize the use of resources.

Electromobility plays a key role in the transition to a more sustainable energy future. However, with the spread of AI-based driver assistance systems, the energy requirements of vehicles continue to rise. GreenAutoML4FAS is therefore focusing on the development of resource-efficient AI for e-mobility. The goal: more safety, comfort and economy without neglecting energy consumption.

Resource-efficient through AutoML

The holistic approach of GreenAutoML4FAS combines hardware, data and model complexity as well as adaptive AI technologies in just one system. Through this interaction, the scientists aim to achieve maximum efficiency that goes far beyond what would be possible with individual optimization of the components. The key to success: the system adapts itself to the diverse tasks of driver assistance systems through automated machine learning (AutoML). The automation of the development process makes it possible to react quickly to new requirements and shorten development cycles. “This innovative approach not only promises to increase efficiency in the automotive sector, but also has the potential to provide impetus across industries,” says Prof. Dr. Marius Lindauer, Head of the Institute for Artificial Intelligence at Leibniz Universität Hannover and consortium leader of the GreenAutoML4FAS project.

PANDA as a trailblazer

PANDA (Platform for the Analysis of Next-gen Driver Assistance) serves as a demonstrator for the practical application and the immense potential of GreenAutoML4FAS. The test vehicle is equipped with state-of-the-art sensors that record and analyze its surroundings in great detail. Board computers optimized as part of the project process the sensor data highly efficiently. “The synergy of optimized hardware, efficient data processing and adaptive software raises the resource efficiency of driver assistance systems to a new level and thus offers a model for future developments in the field of e-mobility,” says Lindauer.

Holistic approach and knowledge transfer

GreenAutoML4FAS is exemplary for a new generation of AI applications that are not only technologically advanced, but also in terms of their resource consumption. The combination of innovative algorithms, efficient hardware and a focused approach to knowledge transfer sets new standards in the development of resource-efficient AI systems. “The publication of the results and findings of GreenAutoML4FAS via open source implementations can give new impetus to the further development of e-mobility. With PANDA as a pioneering demonstrator and the emphasis on the holistic approach, GreenAutoML4FAS is positioning itself at the forefront of efforts to make the future of mobility more sustainable and efficient,” says Lindauer.