If there is a technology that made the early self-driving cars stand out it was the light-detection and ranging (LiDAR) system planted on the top of the car just behind the windscreen.
The array of cameras perched on top of cars like Stanley, Stanford University’s winning entry in the 2007 DARPA Grand Challenge soon moved into more inconspicuous parts of the vehicle. But for a long time, the spinning mirrors needed to bounce laser beams off distant objects and detect their return meant there remain few places to put a LiDAR other than in a big lump on top of the cabin, making it easy to spot the cars run by Waymo and other teams as they roam streets in California.
Even when LiDAR came to production vehicles, such as when Mercedes put it into the late-2021 S-Class, the instrument is still easy to spot, though more carefully camouflaged. Tier-one supplier Valeo managed to pack the electronics and optics into a unit able to sit behind the radiator grille.
Miniaturisation for “LiDAR 2.0” will make it easier to find space for the units but the technology still faces many challenges both inside and outside automotive applications. Though the mechanical construction of older LiDARs seems the most obvious problem to solve, particularly when it comes to size and cost, integrators face major decisions even over what kind of laser the sensors should use for best results.
LiDAR has major advantages over cameras and radar when it comes to some tasks vital to autonomous driving. Its ability to estimate the distance of obstructions, and their overall shape, is much better than that of radar. But what is needed is still a moving target as carmakers grapple with ways to fuse the sensor information from cameras, radar and LiDAR. Changes in one could relax the requirements for other sensors, though this is a delicate balancing act when you consider the wide range of weather conditions in which vehicles must drive.
The LiDAR in the 2021 S-Class was designed for Level 3 driving up to 60km/h. Recently published requirements from the European Union call for greater resolution for vehicles moving at full motorway speeds of 130km/h. Because of the change in braking distances, the sensor needs to be able to reliably detect small details further out. Something like a tyre lying in the middle of a lane may be impossible to recognise reliably at the required detection distance of 160m unless the sensor has a vertical resolution of less than 0.01°. This is because the software needs more than a couple of reflected points to make a judgment.
Addressing adverse conditions
Even with improving resolution, LiDAR has other issues that if not overcome will limit speeds under adverse driving conditions. Zach Bonefas, automation technology leader, at tractor maker John Deere pointed out in a recent online meeting organised by photonics-industry association Optica, how the sub-1000nm near-infrared wavelength range currently favoured for cars cannot readily penetrate the dust clouds that farm equipment frequently throws up. Agricultural machinery at least has the benefit of not having to move very fast.
Fog and road spray present similar issues at these wavelengths because of the way water droplets scatter the light. A study in a fog chamber by engineers at Daimler published in 2019 showed that the effective visibility for the 905nm laser systems can easily be cut to half the measured meteorological visibility. In heavy fog it can worse: with visibility of around 15m, the sensors tested could only detect reliable echoes from objects 4m or less away.
Increasing the intensity of the laser improves effective range but this leads to another problem: eye safety. Though invisible to the eye, the light can still damage the retina and falls under the Class 1 laser regulations. In principle, the longer wavelength of 1550nm is a better choice, not because it is less prone to scattering but because its high-power levels are less harmful, though high powers can still damage the cornea. This wavelength has yet to become a popular choice among carmakers because the systems on the market cost more than the more common 905nm implementations. The increase in output power implied by the need for greater range under adverse weather conditions at higher speeds may push the decision in the direction of the longer wavelength but there may still be a way for 905nm systems to stay within the safety regulations: using lenses to increase the apparent source diameter of the beam.
As with the photonic systems in data communications, temperature is a big problem for stability. And is arguably a bigger problem for automotive systems: the laser transmitters and receivers will either be mounted on the roof or close to the engine, with changes in airflow at different speeds changing how the assembly will be cooled. Thermal drift is already an issue for 905nm lasers but without technological changes, that could get worse at 1550nm, partly because the fibre lasers currently used at this wavelength tend to draw more power than the semiconductor widely available for the higher frequency.
The problem of scanning may prove easier to fix given the number of solid-state designs that have now appeared. To get around the need for a full rotating mirror, MEMS arrays can provide effective beamsteering over wide angles though they cannot compete with the 360° field of view without using several laser and detectors arrays, each covering a field of around 120°.
One approach used by LeddarTech is to mount liquid-crystal modulators in front of the laser elements and switch the pixels between polarisation states to divert the light beam. Three stages of liquid-crystal provide the ability to divert the beam by just over 60°.
Another strategy for a design with no mechanical parts is the flash LiDAR, another technology supported by LeddarTech as well as suppliers such as RoboSense. This emits a broad beam of light in one pulse. Typically, the flash LiDAR has a lower range than the other technologies, seeing the devices used in applications such as toll gates and the automated guided vehicles used in factories as well as automotive designs where speed and range are less important.
Additional features
Though automotive applications face cost and performance challenges, the applications for LiDAR are expanding, partly thanks to the many additional features the technology can bring with smart choices over light sources and optics. For example, the reflected photons contain a large amount of information that can be used to determine much more about a target than its position. Changes to polarisation of the photons produced by coherent sources can pick up differences in materials, such as skin compared with clothing. This points to applications in security and defence. Though something that actively emits near-visible light can give away a user’s position on the battlefield, LiDAR looks as though it could prove useful in combination with radar for detecting lightly buried improvised explosive devices and their control wires at long distance.
Polish supplier Scanway wants to deploy them in space and is working with partners to build a sensor package that will be used to help guide satellites out of the way of space debris that could damage them.
Closer to Earth, the company has developed LiDAR systems to check the quality of wind-turbine blades. And the movie industry has embraced LiDAR as a method of performing motion capture for computer-generated animation and augmented reality that avoids the need to fit actors with uncomfortable emitter arrays. These applications lack the mass-market appeal of automotive and so will have little effect on cost. But they could do much to drive performance.