To err is human
As carmakers and tech companies vie to make better autonomous technology, there already are self-driving trucks and self-driving ride-hailing services.
On the Chinese mainland, internet giant Baidu is working with BMW to develop and test such technology, with testing grounds in Shanghai as well as cities in Anhui and Zhejiang provinces.
Testing self-driving technology under actual road conditions on the mainland is against the law, so companies like Baidu and LeEco have headed to California in the US for road tests.
At this stage, autonomous vehicles are still better suited to the open road than busy urban streets, according to Liu Ming, associate professor at the Robotics Institute at Hong Kong University of Science and Technology, who said none of the models currently on the market are advanced enough to replicate the lightning-quick reflexes and decision-making ability needed for urban driving.
"Technically, it's easier to drive on the highway in driverless mode than to drive at a comparatively low speed in an environment where there are pedestrians, motorbikes and bicycles in front of the car. For this kind of situation, at the moment, it's not easy for (autopilot technology) to deal with," he said.
That milestone requires the evolution of "perception technology", which allows autonomous vehicles to be aware of what is going on around them, he explained. Cameras, radar and lasers act as sensors to detect nearby objects, but judgment about what to do is handled by artificial intelligence.
But Liu said further advancements are needed, as perception technology for self-driving cars has not matured enough to be effective in complex driving environments.
"When a human driver sees the car in front randomly weaving left to right, probably he or she will guess that the driver ahead is drunk or exhausted. The following driver will likely exercise extreme caution," he said. "That extra caution can't be triggered just from direct recognition that there is a car nearby. That judgment comes more from experience, or from other higher levels of understanding in this context."
He predicted it will take four or five years for that kind of technology to mature in autonomous cars. It will entail gathering even more transportation data to improve the judgment of autopilot systems, making it closer to a human's decision-making ability.
On top of all that, the proportion of e-vehicles on the roads will have to be a lot higher before autonomous driving achieves a safe level. Liu cited the example of Hong Kong's complex subway network, the Mass Transit Railway.
Within that network, each train is supplied with data on what is happening with other trains and on the platforms, and other information to ensure the system is safe and efficient.
"If you have all the cars driving on the roads with shared knowledge about what other cars want to do, it'll be much safer," Liu added.
Au believes that driverless cars are the future. But, he said, "I also believe that in such a populous city, the autopilot technology has to be very solid."
