It’s difficult to overlook the rotating sensor atop a Waymo robotaxi. It like a hockey puck that has been bolted to the roof, and in a sense, that is exactly what it is: a mechanical drum filled with revolving mirrors that emit laser pulses in all directions to create a three-dimensional map of the surroundings. It functions. In fact, quite well. However, it looks ridiculous on anything that needs to be sold in a showroom, costs thousands of dollars per unit, and breaks down under constant vibration. That is a serious issue. Because of this, self-driving cars are currently mostly a fleet technology and cannot be purchased.
The attempt to address that is solid-state LiDAR. It creates the same real-time 3D point cloud as mechanical systems by firing millions of infrared laser pulses per second using silicon chips or microelectromechanical systems, which have virtually no moving parts, in place of rotating mirrors and motors. The car is still able to perceive depth, measure the distance to obstacles with centimeter accuracy, and operate in complete darkness and rain, where a camera would become blind. However, the device is much lighter, fits behind a windshield or behind a bumper instead than on the roof, and can someday be produced at the size and cost that consumer cars actually need.
The durability argument is more important than it is commonly acknowledged. It is one thing to have a mechanical sensor on a robotaxi that operates in restricted urban environs and is routinely repaired and replaced. An totally different engineering difficulty is a sensor that must withstand fifteen years of highway vibration, winter road salt, and an adolescent who periodically slams the car door. Unlike spinning mirrors, silicon-based systems do not wander out of calibration. They do not deteriorate. It’s the difference between something that works in a driveway and something that works in a pilot program, so it’s not just a small convenience.
Cost economics are just as significant. Solid-state LiDAR devices around $200 to $500 per at large volume are currently the focus of the most important autonomous programs; this is a significant decrease from the thousands of dollars that defined early mechanical systems. Automakers must commit to incorporating the technology into production vehicles in order to reach that figure. The trajectory is more obvious than it has been in the past ten years, but the chicken-and-egg relationship is still being worked out.
There is no real argument that cameras are no longer necessary. In contrast to LiDAR, visual imaging is capable of reading traffic signals, deciphering lane markings, and identifying faces. However, glare, severe rain, and darkness also work against cameras in ways that LiDAR does not. The greatest autonomous systems make use of both types of sensors, which are complementary rather than antagonistic. The depth perception layer that cameras naturally lack and that the safety case for hands-free driving actually needs is what solid-state LiDAR offers.
It’s still unknown how long it will take for solid-state systems to be produced on a significant scale in consumer automobiles or whether the $200 unit cost target will hold under actual manufacturing conditions. However, the firms pursuing it are well-funded, the engineering approach is logical, and the issue they are addressing is genuine. For more than ten years, autonomous driving has always been five years away. All of that is not resolved by solid-state LiDAR. However, it eliminates one of the more tangible barriers that held it there.
