All companies working in the self-driving vehicle space have a common interest: to convince legislators that the technology is reliable. Microsoft’s recent move to make its latest software platform open source shows that the Redmond giant has really got the message here.
The engineers who nowadays design software programmes for autonomous agents – robots, drones and driverless vehicles – use a similar principle. Machine learning is a technique which enables a software programme to improve itself through experience. Just like human beings, the computer builds up its knowledge from the examples with which it is confronted. This technology for instance recently enabled the AlphaGo artificial intelligence system from Google DeepMind to beat Lee Sedol, one of the world’s best Go players, at his own game.However, just as it takes several years for a human being to learn to move around with perfect ease in the world, a computer needs a huge number of examples, which basically means masses of data, to perform sufficiently well. AlphaGo trained up on tens of millions of Go matches before pulling off its great victory. In similar fashion, Tesla, Google, Uber and the like collect gigantic volumes of data on driving behaviour in order to improve their self-driving software platforms. But if it may be easy to feed a computer with millions of board game moves, it becomes more difficult when it is a case of collecting data out in the wide world.
When it comes to self-driving vehicles, this means testing in real-life situations vehicles which are not yet fine-tuned, with all the risks that this implies.This is where the latest Microsoft creation comes in. The Aerial Informatics and Robotics Platform is intended to generate an ultra-realistic virtual environment where designers and developers can test autonomous navigation software. Cars, drones and robots can all be put to the test on the AirSim virtual simulator. For example, the software transcribes all the small details which constitute a challenge for computers: reflections from the sun, shadows, clouds, and so on. The aim of the simulation is to enable development of software that is sufficiently reliable to be used in the real world but without the risk of running a self-driving car prototype worth several million dollars off the road during testing. The good news is that Microsoft has made the software open source so that anyone can make use of it.
One might reasonably ask why Microsoft is making such a generous gift not only to the general public but to lots of potential competitors as well. Notwithstanding the legendary philanthropy of Bill Gates, the firm’s former CEO, we may venture to doubt that the software giant has made its decision on purely humanitarian grounds. Our first interpretation stems from our earlier remark that when it comes to autonomous systems, data is the nerve-centre in the battle. Microsoft’s software platform certainly has value, but the best way of building a self-driving car today is still to have access to data collected over millions of kilometres by Google, Tesla and Uber, which these firms have been jealously guarding. Google has gathered its data when testing its self-driving cars on fairly empty roads, Uber from trips made by its drivers and Tesla from the autopilot software embedded in its cars. Microsoft’s tool may well help to acquire more data, but it cannot catch up in the blink of an eye on the advances already made by these companies.
A second interpretation is that, for the moment, even though companies developing self-driving systems are competing with each other, they all have a common objective, which encourages them to set aside their rivalries. They all need to convince legislators that their vehicles, drones and robots are safe enough to roam the streets and that the current regulations should therefore be eased. Use of these technologies is still forbidden in the public space and testing is closely constrained by regulation. And without a green light from the regulator, the best innovations in the world are totally useless.
Betting on cooperation
In the United States self-driving cars still cannot be sold on the market, but legislation is adapting as the technology makes progress. A number of states, including California, have authorised testing of autonomous vehicles on public roads, on certain conditions which may vary from state to state. For example, firms wishing to test their vehicles in California have to obtain an Autonomous Vehicle Testing Permit from the California Department of Motor Vehicles, while Florida, Michigan and Arizona have much more flexible legislation, which allows any autonomous vehicle to travel on their roads. Last September, the US government adopted a set of rules designed to act as a framework and promote the development of self-driving cars, while ensuring public safety.
However, four US states – Illinois, Georgia, Maryland and Tennessee – have just passed laws restricting self-drive car testing to bona fide automobile manufacturers. This is bad news for Google and Uber, shows that the tug of war with the lawmakers is far from over and underlines the fact that it is very much in the interests of innovative firms to stand together. The situation is similar in the drone and robot markets. Microsoft‘s initiative marks a step in this direction. Meanwhile, for-profit educational organisation Udacity is now also working on an autonomous car project. Udacity’s co-founder, German entrepreneur educator and computer scientist Sebastian Thrun, who previously headed up development of the Google autonomous car, decided to develop an entirely open system for Udacity’s self-driving vehicles, hoping to speed up development of the technology. “You could say that Google built the iPhone [of self-driving cars] and we’re building Android,” says Thrun, explaining: “Making that software available to everybody reduces the burden for any newcomer to build their own vehicle from scratch, very much like Android accelerated the development of smartphones.”
So, the order of the day now appears to be cooperation first and foremost. Google’s recent face-off with Uber demonstrates however that not all the big players are as yet prepared to take such a philosophical attitude.