- LIDAR allows self-driving systems to see what’s in front of them
- But LIDAR has limitations, too
- Stanford engineers use a clever system to let LIDAR see through a wall
- Their new processing technique is fast enough for self-driving systems
It’s pretty hard to imagine a future for cars that doesn’t involve self-driving systems. Since most accidents are caused by human error, and since even today’s autonomous vehicles are safer drivers than we are, they’ll save thousands of lives. And because traffic congestion is also a result of our poor driving, autonomous vehicles promise to reduce travel times, prevent motorways from becoming carparks, and reduce fuel consumption. All things considered, there’s just too much that’s too good about this innovative tech. It really is the future of driving.
As we’ve written before, the LIDAR (Light Detection And Ranging) systems that allow self-driving vehicles to ‘see’ the road and potential hazards is improving rapidly. Once limited to relatively low speeds, the latest LIDAR can easily handle motorway speeds. That means the tech is almost ready for the road – and the most cutting-edge advances might make autonomous driving almost magically safe.
A team of engineers from Stanford University has just published a research letter in Nature, one of the world’s leading science journals. In it, they describe a potentially LIDAR-compatible system that uses the reflected light from lasers to see around corners and through walls. In the near future, it may also enable self-driving systems to see not only through walls, but through buildings and parked cars as well.
LIDAR allows self-driving systems to see what’s in front of them
LIDAR isn’t new. It came into use shortly after the first lasers. During the 1960’s, for instance, it was used to measure clouds. And by the early 1970’s, Apollo astronauts were using LIDAR to map the moon’s surface. Since then, it’s become pretty common tech for drones and self-driving systems in agriculture. It works by firing pulsed lasers at the objects it needs to see. By measuring the light that bounces back, it takes measurements and constructs a 3D model of what the lasers hit. As its accuracy and speed has increased, it’s become reliable and safe enough to be the main system that lets autonomous cars detect hazards.
![A white self-driving Lexus equipped with lidar tech waiting at an intersection](https://blog.richardvanhooijdonk.com/wp-content/uploads/sites/4/2019/05/Lidar-system-2.jpg)
But LIDAR has limitations, too
But current LIDAR, as good as it is, is limited to line-of-sight. What it can’t see in front of it, it just can’t see. That’s just not good enough for a variety of applications, including self-driving systems. As Matthew O’Toole, David B. Lindell, and Gordon Wetzstein, engineers at Stanford, note, “How to image objects that are hidden from a camera’s view is a problem of fundamental importance to many fields of research, with applications in robotic vision, defence, remote sensing, medical imaging and autonomous vehicles.” In their research letter, they describe a new process for analysing laser light that can solve this problem.
Stanford engineers use a clever system to let LIDAR see through a wall
Here’s how it works. A rapidly firing laser bounces light off a wall, bypassing a partition, and striking a hidden object. As this light hits its target, its photons are scattered, bouncing again and again until a small fraction of them make their way back to where they came from. Rather than measuring light that’s reflected directly – almost none is – O’Toole and his colleagues capture photons that bounce several times during this scattered reflection. As he explains, “We are looking for the second, and third and fourth bounces – they encode the objects that are hidden.”
As Matt Simon reports for Wired, this means that “the researchers are left with … extremely faint traces of light. That’s why they needed a so-called single photon avalanche diode, or SPAD, to make the most of that tiny signal.” As just one of those bouncing, scattered photons hits the SPAD, it begins constructing a 3D model from the data. But to get an accurate picture, the analysis needs to include information like the shape of the wall the laser’s striking and filter out random noise. This isn’t a fast process, but over as little as a few minutes to as much as an hour of measurement, it can provide enough clean data for an accurate 3D picture of the hidden object.
Their new processing technique is fast enough for self-driving systems
The next part is pure high-tech wizardry, or as O’Toole says, “It is almost like magic.” This team tried a new approach to processing and analysis, aiming the sensors at the same point as the laser. The results were nothing short of miraculous. Though they still need a few minutes to gather enough data to analyse, Lindell confirms that processing is virtually instantaneous. “You can push a button on your laptop and process these images in a second …whereas before it took hours on compute-intensive hardware to be able to do this.”
That’s huge. But, yes, technical challenges remain. As O’Toole cautions, “The biggest challenge is the amount of signal lost when light bounces around multiple times … This problem is compounded by the fact that a moving car would need to measure this signal under bright sunlight, at fast rates, and from long range.” And of course, in the real world, you’re not bouncing lasers off smooth, flat walls in controlled conditions. Nor are the lasers used in their experiments the same kind used in current LIDAR systems. But these are challenges that are surmountable using real science and clever engineering. “We believe the computation algorithm is already ready for LIDAR systems … The key question is if the current hardware of LIDAR systems supports this type of imaging,” says O’Toole.
We’re confident that engineers and scientists can make this tech work, and we’d be surprised if something like this system doesn’t make its way into self-driving vehicles in the near future. Detecting potential hazards that aren’t obvious – other vehicles, a boy on a bicycle, a dog chasing a ball between parked cars – can mean the difference between safety and unthinkable tragedy. And we bet that tech, innovation, and a bit of hard work can save lives.