In the 1950s the philosophy informing the futuristic vision of autonomous vehicles was to “make the road guide the car” via electronics and mechanical devices embedded in high tech roadways. Scientists did not seriously consider building smart tech or advanced sensors into the vehicles themselves: it would have been impossible to stuff the 30-ton computers of the time into a Ford sedan; radar and lidar were only seen on advanced fighter aircraft; and the concept of computer vision software had yet to be conceived.
A generation later, attempts at Vehicle-to-Everything (V2X) systems and the US Smart Highway Project of the 1970s and 1980s also failed to gain traction, simply because the required tech had not yet been sufficiently developed.
The machine learning and tech explosion of the last few years has opened up the previously impossible, and self-driving vehicles have arrived. Last month, the Alibaba DAMO Academy and China’s Ministry of Transport Research Institute of Highways (RIOH) launched a new joint lab to tackle the challenge of smartening up the country’s roads.
At the centre of the project is Alibaba’s IntelliSense base station, a hardware device designed to be installed in autonomous vehicles’ operating environment, usually atop streetlights or roadside billboards. Consisting of a variety of sensors and computing units, the base station helps guide self-driving vehicles while also monitoring other vehicles and outside entities. It’s as if there were an attentive traffic cop every two hundred meters along the route.
Synced recently spoke with Gang Wang, who joined Alibaba AI Labs in March and is responsible for the company’s R&D in machine learning, computer vision, and natural language understanding. Wang is a former associate professor at Nanyang Technological University in Singapore and Associate Editor of the IEEE journal Transactions on Pattern Analysis and Machine Intelligence.
Wang says that during the Alibaba team’s testing on multiple open road sections, if the IntelliSense base station was deactivated the self-driving vehicle was unable identify an obstacle such as a pedestrian rushing into the lane at distances less than two meters from the vehicle. Activating the base station enabled the vehicle to sense such obstacles in advance and avoid them successfully.