In Singapore, self-driving development cars are getting a far tougher shakedown than they are in gentle real-world testing in places such as Arizona.
IEEE Spectrum on Monday opened the doors to the Center for Excellence for Testing and Research of AVs, or CETRAN, in Singapore, a 200-acre facility that can replicate producing severe tropical weather. The test center includes nearly everything a self-driving car may find on the real-world roads, albeit with a metal-frame structure that hovers over the car and can produce over 9 inches of rain per hour.
The giant showerhead simulates the torrential downpours that barrel down Singapore and produce flash floods. For self-driving cars, there are no sensors or radar that can handle these conditions today. It’s the CETRAN’s job to train autonomous vehicles for the worst tropical weather.
Niels de Boer, head of CETRAN, said detecting standing water poses enough of a challenge for self-driving cars, let alone determining how deep the puddle is. He added that rain can also interfere with radar and laser signals from lidar, which cuts a self-driving car’s ability to see. Worse, rain can also produce distorted images from cameras, which autonomous cars read as fog.
Although the hardware to overcome these challenges isn’t available yet, developers have already begun machine learning techniques to help cars learn these conditions at CETRAN.
For example, some developers chose fixed points such as tall buildings or poles that even during the heaviest rain can provide the self-driving car with data. Another development team applied a de-hazing filter for camera images to help the self-driving car see better in what the car considered fog. Another used machine learning to fill in the gaps from distorted data points and adjusted the algorithm accordingly.
Government backing has elevated Singapore’s status in the self-driving car development realm as the densely populated country looks to move toward shared vehicles and public transportation.