For decades, only car manufacturers knew the secret ingredients for producing sleek vehicles. But now, the wizards from Silicon Valley are eclipsing powerful engines and sleek bodywork with invisible algorithms.
Until fairy recently, self-driving cars were no more than pie in the sky. But suddenly, everyone is in a rush. Even legislators, who are generally not known for acting quickly. The trouble is, however, that not everyone who’s concerned seems quite ready to go along in full at the same speed. According to the electronica Trend Index 2020, more than 50% of Germans still have reservations about whether they want their car to take full control of the steering wheel in the future. On the other hand, they do seem more prepared to accept advanced driver assistance systems. This is quite an important aspect since, for the first time, we will have to rely in traffic situations on artificial intelligence (AI) for really critical tasks.
After all, in a few years time there will be no cars on the roads without AI, either in full or in part. For car makers, the date is already set – 2020 – although the first step will be high levels of automation rather than completely autonomous driving. At this level 3, cars will indicate automatically, change lanes or keep the car in a straight line, and, if necessary, tell the driver to take over the steering wheel again. It will be many years before level 5 – driverless cars – is reached. At least that’s the general idea. But who knows what AI will come up with in the near future.
AI for autonomous driving
Developers are already rubbing their eyes. Having been unsuccessful for many years, artificial intelligence, or rather deep learning – a revival of artificial neural networks – is now racking up one success after another. This has become possible as a result of enormous computing power and vast amounts of digital data. After all, the system, comprising several levels of neural networks, has to be fed, for example, with millions of tree images. To put it simply, like a baby it gradually learns how to identify trees as such.
As well as the exploding volume of “training material” the breakthrough came with the unbelievable computing power of the CUDA graphic cards from NVIDIA at the end of the 2000’s. While Intel processors are designed universally and can calculate almost everything with a few cores, in graphics processors thousands of cores share an enormous number of simple computing operations that are always the same. Ideal for scientific calculations, weather simulations, and also for neural networks. This is because, basically, they learn by simple multiplications, or what are known as convolutions.
This explains why NVIDIA is also very popular among the “self-drivers”. Just recently, the world’s biggest supplier to the automotive industry, Bosch, announced that it would develop artificial intelligence for semi and fully autonomous driving by 2020 with the California-based company using NVIDIA’s Deep Learning software and the NVIDIA hardware platform. Bosch plans to invest about €300 million in a Center for Artificial Intelligence (BCAI). “No Bosch product without artificial intelligence in ten years” is the company’s plan for the future.
With the next generation of NVIDIA DRIVE PX boards, self-driving cars could also reach level 4 (full automation). At this level, the system completely takes over the “steering wheel”. The driver is nevertheless still “available” in an emergency.
Necessity is the mother of cooperation
Apart from Bosch, the ZF Group is also relying on NVIDIA and HERE, the mapping company owned by Daimler, Audi, BMW, and, of late, Intel plans to extend its HD card material with NVIDIA AI features. Highly precise HD cards that are always up to date are a requirement for autonomous cars.
NVIDIA is actually a direct competitor of Intel, which, after its recent acquisition of the former Tesla partner Mobileye, has become a key player in the field of self-driving cars. And, of course, with modern versions of their processors, both Intel and ARM are looking at artificial intelligence and machine learning.
The development of autonomous driving cars is so complex and cost-intensive that success is possible only through major cross-manufacturer and cross-industry alliances. And the market seems to be big enough for all of them. But there are vast differences when it comes to the figures, not least because the technologies for self-driving cars are extremely fragmented. For example, Stratistics MRC estimates that the global ADAS (Global Advanced Driver Assistance Systems) market of USD22.52 billion in 2015 will rise to almost USD90 billion by 2022. MarketsandMarkets forecasts slightly more than USD 40 billion by 2021.
A recent study from KPMG shows why everyone has to jump on the “autonomous” bandwagon. According to the leading audit services and consulting company, by 2030 buyers can expect to see about USD 1 trillion worth of autonomous mobility products and services. A cake that could be vital for the survival of car manufacturers. After all – who knows how future generations will want to get from A to B.
KPMG: Global Automotive Executive Survey (pdf, 56 pages).
IHS Whitepaper: Artificial Intelligence in Automotives
Stanford University: Artificial Intelligence and Life in 2030 (pdf, 52 pages).