Q1 2020
The units visible on many autonomous vehicles from the likes of Waymo are a form of electromechanical scanning LIDAR and can cost tens of thousands of dollars with the cheapest variants bottoming out around 6000 dollars each. This is due to the operating principle of this type of LIDAR which requires spinning optics (mirror, lenses) and electronics (lasers, sensors) with high precision. To communicate with the electronics in the spinning assembly requires the use of component called a slip ring. Manufacturing to the fine tolerances required for the rotating assembly and making the slip ring reliable are already expensive, making these units robust enough to survive the shocks of on-highway use is a very expensive proposition. Though expensive, companies gladly pay these sums as there has been no other form of LIDAR that can match its capability.
In a bid to dethrone the reigning champs, a number of technologies have cropped up to both surpass the capability and remedy the major manufacturing costs of traditional electromechanical scanning LIDAR. Most notable are a group known as solid-state LIDAR. As the name implies, there are no moving parts allowing it to be smaller, extremely robust, and significantly cheaper to produce.
There are a few competing solid-state technologies, the most prominent being flash and phased array. Flash LIDAR operates similar to a standard flash camera, but calculates the distance from the flash bouncing off an object. This requires a powerful flash that can be damaging to human retinas, so either exotic materials or reduced range must be used to turn down the flash power. Phased arrays operates similar to a radar array, but uses light and many small optical antennas to detect the pattern from a precise burst. While range is excellent at eye-safe radiation levels, this technology has been plagued with development issues leading to delays in mass production for years.
What should be clear so far is that there while LIDAR is and impressive and revolutionary technology in computer vision, it is in no way mature. In many ways, the LIDAR industry looks very similar to the digital camera industry a few decades ago. For example, the sensors available today frequently have a low field of view and/or poor resolution just like the early digital camera examples. The unique aspect of being an active sensor, however, may mean that there will be more relative difficulty in producing more capable units in terms of range.
Long-range (typically 100m+) LIDAR units, are in high demand by automotive OEMs and large research groups like Waymo. Just about every discussion with suppliers collapses when they want to know which research institution this work is associated with or how many units per year are desired. Other companies have been promising that sample units will be available next quarter or two with mass production starting another quarter or two after that since about 2016. Either way, there are very few publicly purchasable 3D LIDAR sensors. It seems only a couple of Chinese companies selling new units and some in the used market, though the lifetime of those units may be suspect.
As people will still be driving cars for quite some time, these light-based sensors must be eye safe. Practically all commercial LIDAR use infrared (IR) light due to the retina's reduced ability to absorb those wavelengths allowing for higher power lasers while remaining eye safe. Most units operate in the near-IR band typically within 800-1000nm (commonly at 850nm and 905nm) with a smaller number operating in the short-wavelength IR band at 1550nm. The human eye absorbs significantly less light at 1550nm than the NIR band allowing significantly higher laser power usage but the light is more readily absorbed by water vapor. This leads to some performance gains in good conditions with performance reductions in higher humidity and precipitation while incurring a significant increase in power consumption.
Another material that absorbs IR light very well is asphalt, the predominant visible ingredient in pavement used as a roadway surface. The amount of light reflected ranges from 5% to 20% no wear to exposed aggregate, respectively (This will eventually be compiled into a report). This low reflectivity drastically reduces the effective range of LIDAR. Most units list their effective range at a target with 80% reflectivity and some also include a range with a 10% reflectivity target. From these units it appears that about 100m of range on an 80% target results in about 30m on a 10% target. This reduced range is important for not only detecting asphalt pavement but other low reflectivity objects like dark car paint and dark clothing on pedestrians.