Friedrichshafen/Berkeley, California. ZF, one of the world’s largest suppliers to the automotive industry, has entered into another pioneering strategic research partnership to coincide with the opening of its Innovation Hub in Silicon Valley. Together with researchers from the renowned University of California, Berkeley, the company will in future harness the self-learning machines that are crucial to fully autonomous driving in the automotive field.
When people are faced with complex traffic situations, their intuition often helps them to make the right decisions. For example, experienced drivers may know without clear indications that the car in front is about to pull out and overtake, or whether a pedestrian is about step out onto the road. Transferring this ability to computers in a highly complex automotive environment in order to establish one of the key requirements for fully autonomous driving is one of the biggest challenges currently facing the automotive industry. “Our latest research collaboration will significantly boost our Vision Zero Ecosystem in two areas that are key to fully autonomous driving – computer vision and deep learning,” says Dr. Stefan Sommer, CEO of ZF Friedrichshafen AG. The researchers at the University of California, Berkeley are among the world’s leading minds in both fields. In 2017, Times Higher Education placed the university, which currently has around 37,000 students, sixth in its World University Rankings. With the newly founded Berkeley DeepDrive (BDD) Center, the university has teamed up with high-caliber industrial partners such as ZF to create the ideal conditions for achieving the ambitious aims. The BDD consortium brings together faculties and researchers from several departments and centers to combine the latest technology with real applications in the automotive industry. “Even though dramatic progress has been achieved in fields such as computer vision in many areas of industry in recent years, the applications are yet to reach the automotive industry. We now wish to change this,” says Professor Trevor Darrell, head of the multidisciplinary center.
The aims: To make a quantum leap forward in terms of awareness of surroundings, learning ability, and development pace and safety
In terms of autonomous driving, the highly complex environment in which automobiles move around together with numerous other road users places the highest possible demands on the system algorithms. To make matters more difficult, even though the normal test runs used until now have covered millions of kilometers, they still don’t come anywhere close to covering all the conceivable traffic events or risk situations. Different ways must therefore be found to ensure the error-free functionality of the system. Part one of the strategy involves algorithms being capable of optimizing themselves in future by means of machine learning. To this end, machine learning uses so-called neural networks modelled on the way in which the human brain works. If this involves particularly complex neural networks with many hidden layers, it is known as deep learning. Part two of the strategy involves vehicles equipped with the necessary sensors learning how to achieve ever better outcomes using the incoming data. Individual system adjustments are aggregated in the cloud, optimized once again, and re-sent to the entire vehicle fleet. In this way, both the development pace and quality can be increased on a lasting basis.