Smart computers ("artificial intelligence") can be used for many purposes. They can help in (financial) planning for companies, trace fraud and assist doctors in diagnosis. However, the most promising use of intelligent computers is in self-driving cars. That's why developers of smart computers, scientists and carmakers meet at GTC 2018.
One year ago Autozine was also present at GTC and at that edition emphasis was on development in hardware. The organiser as well as the leading manufacturer of hardware for artificial intelligence, nVidia, reported breakthrough results back then. Computers to be installed in a self-driving car no longer took up an entire boot. Instead, a computer smaller than a laptop and more energy efficient than a car stereo could do the job. Together with even faster servers for developers, all steps in the field of hardware were taken.
The most important points remaining in 2017 were acceptance from consumers, permission from the law and initiative from the industry. On all these fields a lot of progress has been made in 2018, starting with the industry.
A taxi ride is expensive because of the driver. But a ride with a robot taxi will be even cheaper than driving your own car. Sales of privately owned cars will therefore decrease significantly in the future. Up to now, this was a reason for many stakeholders to delay the self-driving car as much as they could. Companies like Uber, Google and Apple soon realised that things are in fact very different.
According to research by nVidia, 100 million cars are sold each year, while 10 trillion kilometres are being travelled. If carmakers start selling kilometres travelled instead of cars, their profits will increase exponentially. Since carmakers started realising this, the race for the self-driving car has taken off between traditional carmakers and tech companies.
To stimulate the industry even more, nVidia now offers ready-made "do it yourself" kits which can turn every car into a self-driving car with relative ease. All the necessary sensors, computer and even software are now available to even the smallest carmakers. Now it's up to the car brands to improve on this, possibly with the help of the many software developers at GTC 2018.
At the same time, a trend is visible where not only privately owned cars, but also utility vehicles are turned into autonomous vehicles. For example, "Enway" demonstrates a self-driving street sweeper. It solves two problems: it's hard to find drivers for these machines and a robot sweeper can also work at night (don't worry: the new street sweepers are electric vehicles and don't produce as much noise as the diesels that now wake up entire neighbourhoods).
A traditional car has to meet a certain number of requirements in order to be allowed on public roads. These criteria have to do with safety, construction and emissions. They are simple rules and the answerers can be expressed in numbers or true/false answers. Up to now it was much harder to allow a self-driving car on the roads, because the so-called "neural network" in the brain of such a car can't be examined as easily.
The German "TÜV SÜD" now takes the lead, as it developed a series of tests which self-driving cars have to pass. If an artificially intelligent system passes these tests in a simulator, it receives a "licence" to make it road legal. Theoretically, this makes it possible for the first self-driving cars to be sold to consumers. American government subsidiaries are working on similar tests as well.
Self-driving cars have been being tested for many years. At GTC 2018 several autonomous cars demonstrated their capabilities and all went well. However, we're still waiting for the final breakthrough because self-driving cars shouldn't just do well, they have to do perfectly. Statistically, self-driving cars already make less mistakes than human drivers, but consumers only allow the electronics to take control when computers make zero mistakes.
The many software developers at GTC 2018 showed how the last problems will be solved. For example: artificially intelligent systems from Perceptive Automata don't just recognise pedestrians, they can read body language and predict the intention of a person. Does someone walk up to a zebra crossing in big determined steps? Or is a person wondering around, on the phone, forgetting about traffic altogether? Just like a human driver, a computer can now determine what pedestrians will do.
nVidia has developed even faster chips for self driving-cars than last year, so that the computing power keeps up with the even more advanced software. In the near future the new "DRIVE AGX Xavier" will be deployed to improve the current self-driving functions (look ahead with the driver, make small steering corrections and brake when necessary), making them much more reliable. Daimler (Mercedes-Benz and Smart) and Bosch are working with these new computers, but haven't announced actual products implementing them yet. Volvo takes the lead and announced at GTC 2018 that as of now the latest generation of nVidia hardware will be in all future models.
By making existing driver's aids ("level 2") more reliable by using "DRIVE AGX Xavier", consumers will trust them more. The technology will not only be more reliable over time, it will also be able to handle more complex situations. Gradually, the computer will assist the driver more and more, until the car finally drives all by itself.
Because robot taxis work in a limited area, they are easier to develop than self-driving privately owned cars which have the function everywhere. That's why nVidia CEO Jensen Huang expects the first test with full self-driving taxis to be early 2019. In 2020 he already expects tens of thousands of robot taxis worldwide.
Huang is convinced that within 10 years all vehicles, from private cars via taxis to lorries, will drive autonomously. This means the end of traffic jams, pollution and traffic accidents and that's what it's all about at GTC 2018.
Development in the car industry goes step by step. Progress is just enough to be able to bring a slightly better car to market every time and give consumers a reason to upgrade. Progress is slow enough to keep development costs limited. It's an economic balance: progress can be much quicker, but in the long run this is less profitable.
In the world of artificial intelligence progress knows only one speed: full throttle! This is also driven by economics: the one developing the first self-driving car can potentially change the world and lead the industry. Software houses therefore develop more advanced applications as quickly as they can. Market leader and hardware maker nVidia offers ever faster computing power, to make sure the ever more complicated software will run effortlessly. Also, nVidia now makes this power available to a larger market by offering ready-made development kits for self-driving cars.
In the time frame of just one year the self-driving car has come much closer. Because the required hard and software is now easily available, even smaller carmakers are developing their self-driving vehicles. Governments have taken significant steps by issuing the first "drivers licences" for self-driving cars. To gain confidence from consumers, the current assistants will soon become more reliable. Once everyone trusts the simple assistant, drivers will gradually hand over the steering wheel to the computer.