Many autonomous vehicles use technology known as LIDAR (light Imaging, detection and ranging) to understand the world around them. In the simplest terms, LIDAR bounces light off objects to see where they are and to build a picture of what they look like. With this information, the car can then differentiate between various objects including vehicles, road markings and people to safely navigate the roads.
Naturally, you’d assume such technology depends on the objects being within the line of sight, but what if autonomous vehicles could also build a picture of objects that aren’t immediately visible? It might sound like something out of a science fiction novel, but researchers from Stanford University have developed highly sensitive laser technology that can see things around corners by reflecting light off nearby objects.
The team, whose research is published today in the journal Nature, placed a laser alongside a highly sensitive light detector. After shooting pulses of light at a wall, the researchers found those pulses would bounce off objects around the corner and then bounce back to the wall and indeed the detector.
The speed at which the sensor can scan the image is one of the main obstacles to the technology’s practical implementation. In the research paper, it is quoted as taking between two minutes and an hour, depending on light conditions and the reflectivity of the hidden objects, which is obviously too slow for it to be of use in the real world.
Graduate student David Lindell and post-doctoral scholar Matt O’Toole working in the lab
But that’s getting better. “Since the paper was accepted, we have been able to get the acquisition time down to two frames per second and we can further accelerate it by using a more powerful laser,” senior author Gordon Wetzstein told us. “That would bring us into the real-time domain, which would be useful for autonomous driving.”
It isn’t the first study to explore the idea of bouncing lasers around corners to build pictures of objects, but where it undoubtedly breaks new ground is in the development of an algorithm that can process the scans in a matter of seconds on any ordinary computer.
“We have found a way to do this in a very memory-efficient, computationally efficient way that drastically lowers the amount of resources that [are] required to perform this type of computation. We [went] from basically hours to seconds”, reveals co-lead author of the paper Matthew O’Toole.
Furthermore, the researchers believe their algorithm is compatible with current LIDAR systems. Besides testing it with the variability of the real world, the main question is how the technology might be made to work with the hardware in LIDAR systems, but Wetzstein seems confident on this too.
“The hardware solutions we have developed [are] compatible with existing LIDAR systems, Wetzstein tells us. “Our methods could potentially be implemented with commercially available systems with the right changes to their software.”