EXPERIMENTAL self-driving cars continue to make regular forays onto the roads. After a trial in Pittsburgh, Uber, a taxi-hailing-app company, launched several of its “autonomous” vehicles onto the streets of San Francisco on December 14th—and promptly ran into a row with officials for not obtaining an operating permit, which Uber insists is unnecessary as the vehicles have a backup driver to take over if something goes wrong. General Motors said it would begin testing self-driving cars in Michigan. For these and other trials one thing is essential: providing the vehicles with a reliable form of vision.
As no man-made system can yet match a pair of human eyes and the image-processing power of a brain, compromises have to be made. This is why engineers use a belt-and-braces approach in equipping vehicles with sensors that can scan the road ahead. That way, just as your trousers will stay up if one or other of belt and braces fails, if one system misses a potential hazard, such as an oncoming car or a pedestrian, the others might spot it and direct the car to take evasive action.
Three of the sensory systems currently in use in autonomous vehicles—cameras, ultrasonic detectors and radar—are reasonably cheap and easy to deploy. A fourth, lidar, is not. Lidar employs laser scanning and ranging to build up a detailed three-dimensional image of a vehicle’s surroundings. That is useful stuff as the lidar image can be compared with the data being captured by the other sensors. The problems are that lidar is bulky (it hides in the roof domes of Google’s self-driving cars and, as pictured above, in the revolving beacons that adorn Uber’s vehicles), mechanically complicated and can cost as much as the unadorned car itself.
Smaller, cheaper lidars are being developed. One of the most promising comes in the minuscule form of a silicon chip. Prototypes have been delivered to several big automotive-component suppliers, including Delphi and ZF. If all goes well, within three years or so lidar chips should start popping up in vehicles.
A chip off the old block
The company bringing these miniature lidars to market is Infineon, a German chipmaker. This firm is one of the biggest producers of the chips used in radar detectors. Radar works by sending out radio pulses and detecting the reflected signals that have bounced off objects. The time delay between emitting a pulse and noting its reflection is used to calculate how far away the reflecting object is. If that object is moving, then its speed can also be determined. This determination comes from a slight shift in the frequency of the reflected signal, caused by the Doppler effect (the phenomenon that also causes a passing fire-engine’s siren to change pitch).
Around 15 years ago radar sensors were specialised pieces of kit and cost around $3,000. Infineon found a way to make them using a standard silicon-based manufacturing process and, by integrating many of the functions of a radar onto a single chip, boost performance. That has brought the price down to a few hundred dollars. As a result, radar chips have become an essential part of an autonomous car and are increasingly used in conventional vehicles too, to provide safety features such as automatic emergency braking.
The race is now on to shrink lidar in a similar way. Lidar was developed as a surveying method following the invention of the laser in the 1960s. It employs a laser beam to scan an area and then analyses the reflections that bounce back. As light has a much shorter wavelength than radio waves do, it is more readily reflected from small objects that radar might miss. Lidar is used to make maps, measure atmospheric conditions and by police forces to scan accident and crime scenes.
Typically, a lidar employs revolving mirrors to direct its laser beam, which is usually in the invisible near-infrared part of the spectrum, rather than the visible part. Commercial lidar can cost $50,000 or so a pop, but smaller, lower-powered versions are now available for $10,000 or less. A number of lidar makers, such as Velodyne, a Californian firm, are trying to develop what they call “solid-state” lidars, which are miniaturised versions with no moving parts. Some researchers are using a flash of laser light instead of a beam, and capturing the reflections with an array of tiny sensors on a chip.
Infineon, however, has taken a different tack and is using a micro-electro-mechanical system (MEMS). This particular MEMS was invented by Innoluce, a Dutch firm which Infineon bought in October 2016. The device consists of an oval-shaped mirror, just 3mm by 4mm, contained on a bed of silicon. The mirror is connected to actuators that use electrical resonance to make it oscillate from side to side, changing the direction of the laser beam it is reflecting. This, says Infineon, permits the full power of the laser to be used for scanning instead of its light being dispersed, as it would be in a flash-based system.
The MEMS lidar can scan up to 5,000 data points from a scene every second, and has a range of 250 metres, says Ralf Bornefeld, Infineon’s head of automotive sense and control. Despite its moving mirror, he thinks it should prove as robust and reliable as any other silicon chip. In mass production and attached to, say, a windscreen, the MEMS lidar is expected to cost a carmaker less than $250. These tiny lidars would have other applications, too—in robots and drones, for example.
Many engineers, Mr Bornefeld included, reckon autonomous cars of the future will use multiple miniature lidars, radars, ultrasonic sensors and digital cameras. Each system of sensors has advantages and disadvantages, he says. Combining them will provide a “safety cocoon” around an autonomous vehicle.
Radar measures distance and speed precisely, and works in the dark and in fog—conditions in which cameras might struggle—but the images it yields can be difficult to classify. Moreover, some materials (rubber, for example) do not reflect radar waves well, so radar c
ould have difficulty noticing, say, a dangerous chunk of tyre from a blowout lying in the road. With good visibility, the car’s cameras should spot the bits of tyre. The cameras capture high-resolution pictures, use artificial-intelligence software to analyse them, and then apply image-recognition techniques to identify objects that need to be avoided. Lidar, with its ability to build detailed images of even small objects and operate in the dark, should spot the tyre, though it, too, might struggle to do so in dense fog. Ultrasonic detectors, meanwhile, will continue to play a part. They have been around for a while and work in a similar way to radar, but instead use high-frequency sound inaudible to humans. They would not see the tyre chunk—at least, not until too late—for they usually lack the range. But they are cheap and make excellent parking sensors.
Google, Uber and most carmakers who aspire to make autonomous vehicles already use lidar. They ought, therefore, to welcome its miniaturisation with open arms. But not everyone is convinced of lidar’s worth. Elon Musk, the boss of Tesla, a firm that makes
electric cars, has spurned the technology. He has said the camera, radar and ultrasonic systems that provide the Autopilot autonomous-driving mode in Tesla’s vehicles are improving rapidly and will be all that is necessary.
Mr Musk may, though, change his mind. In Florida, in May 2016, the driver of a Tesla using Autopilot at high speed was killed in a collision with a lorry turning across the road in front of him. Although Autopilot users are supposed to keep their hands on the wheel and their eyes on the road (just as, for now, the backup drivers in Google and Uber cars do), it appears the Tesla’s cameras and radar either failed to spot the lorry—which was painted white and set against a brightly lit sky—or thought it was something else, such as an overhead sign. Whether lidar would have made the correct call, as some think it would, no one will ever know. But when more driverless cars venture onto the roads in earnest, having plenty of belts and braces might help reassure their passengers