Autonomous cars, artificial intelligence and the future of search

The future of driving will be as easy as riding a horse which knows where its going and how to get there, according to Chinese web giant Baidu’s top artificial-intelligence (AI) researcher.

Self-driving cars will understand the road ahead of them and autonomously navigate around pedestrians and other obstructions, said Yu Kai, head of the Baidu Institute of Deep Learning (IDL).

Based in Beijing, IDL is often known as “Baidu’s Brain”, a place for the search engine powerhouse to bring together the smartest scientists from all over the world to work on innovations and new technology.

“We’ve been busy working on the autonomous car, and we aim to put a prototype on a Beijing highway in the second half of this year,” Yu said in a recent interview with the South China Morning Post in the firm’s headquarters.

“Our idea is not that a car should totally replace the driver, but that it will give the driver freedom. So the car is intelligent enough to operate by itself, like a horse, and make decisions depending on different road situations,” said Yu, who is also senior director of the company’s image search department.

“But whenever the driver wants to resume control, you can do that,” he said, adding that the Baidu self-driving car would have a steering wheel as well as gas and brake pedals.

Yu explained that a self-driving car will require a very accurate on-board map to be able to navigate effectively.

“And it has to stay within a 10-20 cm margin of its intended route. Simply letting the car run on a chosen road or street isn’t good enough, we have to have control over which lane it runs on,” he said.

Since finishing its incident simulation system, and with a high-precision mapping system almost complete, Yu said the team at IDL has begun testing the vehicle in their garage, seeing how it reacts to tasks like parking, or to sudden stops.

More tuning needs to be done before the self-driving car will be road ready however, as the team is still looking at ways to improve the vehicle’s algorithms in situations when, for example, there are obstacles in front of it that it needs to automatically avoid, or when a car in front of it changes lanes unexpectedly.

Yu said that one arduous part of the team’s work is incorporating multiple different sensors and recognition technologies into a single machine. As an example, he explained that the technology that enables the car to identify and avoid obstacles is completely different from that which can read speed limit signs.

The project puts Baidu, which has been researching its self-driving car in collaboration with automaker partners such as BMW, in competition with Google and Apple, both of which have announced their own smart car projects.

In December, Google unveiled a fully functioning prototype of their driverless car and unlike its competitors, the company has been conducting live road tests for a number of years.

Mark Reuss, product-development chief at General Motors, told Bloomberg in 2014 that Google could soon become a “serious competitive threat” to the US auto industry.

While the technology is often referred to as “driverless cars”, Yu dislikes the term, preferring to call Baidu’s system “autonomous driving” because of the difficulties of having a truly driverless vehicle, especially in a potentially-dangerous city setting.

“We will see many autonomous cars on the road in five years, and true driverless cars further down the road,” he said.

And before the technology can mature to such a point there are many obstacles to overcome such as how the car handles differing climates and pollution levels. Government regulation is also a potential risk to the project, though Yu hoped it would promote safety without stifling innovation.

The development of the car has opened a world of opportunities for Baidu, including getting more people to use its mapping services, which can help drive adoption of offline services, said Kaiser Kuo, the firm’s head of international communications.

Baidu is not the only mainland Chinese internet company looking to get into the auto market. LeTV, a maker of internet-enabled televisions, has said it will invest billions of dollars in developing a connected electric car with BAIC Motor. Last month, e-commerce giant Alibaba also announced a fund to promote internet connected automobile technology in collaboration with SAIC Motor.

“In the age of the internet economy, cross-boundary integration has become an inevitable trend,” SAIC said in a statement.

“The cars of the future must be Internet-oriented.”

Baidu has an advantage in that it has already invested heavily in deep-learning, which not only powers its autonomous vehicle project, but also its search, advertising, and speech and image recognition technologies.

Deep learning, also known as deep structured learning or hierarchical learning, is a branch of machine learning and basic artificial intelligence in which a set of algorithms are used to model complex situations and abstractions in data using model architectures, composed of multiple non-linear transformations.

The technology is used in Baidu’s advertising system to measure what qualities of an ad make people click on it, select ads based on those criteria, and serve them at opportune moments, allowing Baidu to charge higher prices than its competitors. Baidu chief executive Robin Li Yanhong told investors in April that the technology had helped lift first-quarter profits and revenues.

Baidu had a 82 per cent share of mainland China’s 59 billion-yuan search market last year, technology consultancy iResearch said in a report in February.

Yu said it has also led to a double-digit percentage growth in terms of advertising revenue: “We have seen obvious improvements in click-through rates after we implemented deep learning.”

Research company eMarketer said in a recent report that Baidu’s search advertising revenue was predicted to rise by 34 per cent to US$7.1 billion this year, up from US$5.3 billion in 2014.

Deep learning has also helped lift Baidu’s speech recognition service to being the best in the world in terms of recognising Chinese speech, Yu said, adding that significant advancements have also been made in terms of optical character recognition, face-recognition, and generic object recognition.

“In all these areas, we’re either world leading or one of the top,” he said.

In January 2014, Baidu Translate launched a feature that can, in seconds, identify an object in a photo and identify it in written and spoken English or Chinese. Though notable failures of the technology were quickly shared online (such as when it identified a urinal as a “European style chandelier”), the technology has continued to improve and become more accurate.

Image recognition technology is also used in search, allowing users to snap photos of things in front of them rather than type queries.

In February , Baidu launched the IDL-developed StockMaster app, which uses artificial intelligence to predict how stocks, sectors and markets may perform. The app uses artificial to analyse news, markets and Baidu’s own search engine data, providing predictions of upcoming changes. It does not recommend stocks however.

Deep learning has the potential to be applied in many more areas than it currently is, Yu said.

Healthcare is a key sector that could benefit from artificial intelligence: “If a machine could help us read all medical records and new lab findings and understand them, and put them into a huge knowledge base, we could discover insights currently unavailable to us because we can’t read so much literature.”

“The computer will learn for us,” he said.

Full article available at South China Morning Post