Self-driving car tech ready to hit the street
Valeo's AI-based system works '99% of the time' says innovation manager
TALLULAH LUTKIN and TOGO SHIRAISHI, Nikkei staff writers
PARIS -- Self-driving car technology can now adapt to almost all driving situations, including motorways and parking, said Guillaume Devauchelle, vice-president of Group Innovation and Scientific Development at Valeo, one of the world's biggest automotive suppliers, based in France. But some regulatory questions -- such as whether artificial intelligence can be taught to violate traffic regulations if necessary -- still have to be addressed by authorities.
"It works 99% of the time," said Devauchelle, in charge of self-driving car technology at Valeo, in an interview with the Nikkei Asian Review. Valeo develops the different sensors needed for self-driving cars: cameras, scanners, radars, ultrasonic sensors, and so on.
"At the moment, we've achieved that threshold of quality on motorways, for example," said Devauchelle. Valeo's prototype has already circled the Parisian ring-road for 24 hours in self-driving mode 99% of the time. The company will soon start tests on the busy streets of Paris itself.
Another important feature Devauchelle emphasized and which will soon be on the market is low-speed maneuvers. "We can do completely autonomous parking maneuvers; the car can park itself without a driver inside," said Devauchelle. This technology uses mainly ultrasound sensors and is now fully functional. Valeo also helped develop a low-speed driverless shuttle bus which is already driving around La Defense, a pedestrian business district in Paris.
Devauchelle also stressed the company's development of AI. Valeo's AI uses deep learning, meaning the car improves its knowledge of the road and its driving skills the more it drives, and can also adapt to the characteristics of each driver. "The behavior of the car will evolve over time," making the car safer to use the more it drives. Valeo recently announced it was launching Valeo.ai, the first global research center in AI and deep learning, based in Paris, which will host AI specialists and focus on applications in the automotive industry.
Some finer points, however, still need to be resolved. The sensors do not work in every situation -- for example, they are not always reliable when coming over a hill in the sunset or if the road is covered in snow. What's more, different countries have different road conditions, which need to be taken into account. "In Las Vegas, it's too hot to paint white lines, so they use bronze studs which are harder to see for the car," said Devauchelle, citing just one example of the many local characteristics they have come up against.
There are other issues which go beyond the car's safety record. Valeo wants its features to work everywhere, but there are some cases where strictly abiding by the highway code would not be the right thing to do. Devauchelle gives the example of the infamous Place de l'Etoile in Paris, where drivers have to break right-of-way rules if they want to cross the huge roundabout. "We want to be able to do it, but as of today it's not strictly legal," said Devauchelle. "The issue goes beyond Valeo. It's an issue for the manufacturer and a societal issue."
Devauchelle is convinced that self-driving cars will ultimately be safer than human drivers, because the car is constantly focused on the road and gathering data, unlike a human driver, who can be distracted. "If you've made the right decisions, accidents won't happen," he said. "Nowadays we manage what happens after the accident: airbags, passenger security features, etc. But the goal is to prevent the accident from happening in the first place." Valeo has set itself a very stringent safety objective: one fault per billion kilometers traveled, which is much lower than with human drivers.
The market for self-driving cars has huge potential. Even though fully self-driving cars are currently banned everywhere in the world, a Boston Consulting Group analysis estimates that 12 million fully self-driving cars are expected to be sold worldwide every year by 2035, with self-driving features capturing 25% of the new car market by that time. Valeo's direct competitors are other automotive suppliers such as Continental or Bosch, but tech companies and car manufacturers also want a share of this lucrative market. To remain competitive, Valeo is teaming up with innovative start-ups such as CloudMade, which will help improve data storage for AI features. Valeo also bought the German electronic company Peiker in 2016, which will contribute to reinforcing the cars' cyber-security features.
Valeo has seen healthy growth in profit, which has risen by 25% per year on average for the last five years. It supplies manufacturers all over the world, but is keen to penetrate the Chinese market. "We have to aim for China because it's half the market," said Devauchelle, though at the moment, Asia only accounts for 27% of Valeo's sales.