Self-driving may have slipped below the radar during lockdown, not helped by a spate of earlier crashes and reality checks, but as in other sectors advances already in train are still proceeding. Volvo is the latest to keep momentum going by announcing that cars it makes will incorporate Lidar (light detection and ranging) systems from 2023.
This follows a deal with Swedish start-up Luminar in which it invested in 2018 with a view to incorporating that firm’s Lidar sensors, recognized as among the most accurate available at an affordable price. The Lidar sensors will be available in 2022 as part of Volvo’s SPA 2 architecture, the successor to the SPA 1 framework of most Volvo cars on the road today.
Lidar is arguably the most important of the sensing technologies for autonomous driving, illuminating targets with laser light pulses and analyzing the reflection observed by a sensor. Differences in laser return times and also wavelengths are then applied to detect obstacles and their distances, enabling near real time 3D representations of the changing immediate environment to be maintained.
Several key developments have made Lidar more applicable to terrestrial applications such as self-driving, as opposed to airborne ones. One is the use of indium gallium arsenide (InGaAs) chips rather than conventional silicon in detectors, which allows less hazardous infrared laser pulses to be used. Another is the move away from mechanical components to aim the chips correctly, which are a cause of unreliability and failure.
The significance of the Volvo move is that it will be one of the first automotive makers to incorporate advanced Lidar in a standard car, taking advantage of rapid falls in price of the technology. This is where Luminar has scored, not by being quite at the cutting edge of Lidar technology but by bringing the price of the sensors down below $1000, when the leading vendor Velodyne had been charging as much as $75,000 for an admittedly top end version with full 360 degree field of vision.
That means the technology previously confined to the highest end vehicles from the likes of Waymo and more recently Tesla for example can now go into midrange cars that will offer a degree of autonomy.
Some context is required here to put Lidar in perspective and explain why recent arguments within the industry over its role are red herrings. The first point is that there is a huge difference between deploying self-driving cars within controlled environments such as industrial campuses, or even for platooning where vehicles travel in convoy, and completely ubiquitous autonomy extending to all roads and all vehicles under all road conditions.
That is because there are so many exception or edge cases that will have to be accommodated for full autonomous driving everywhere. That will take years of testing and piloting to tease out and address to the full satisfaction of regulators as well as the public. It is worth noting that even Level 5 driving, the highest degree of autonomy currently specified, suggests that such vehicles may be prohibited under certain extreme conditions, perhaps including heavy fog. Of course human driving is also dangerous under such conditions but not normally prohibited.
The other key point is that autonomous driving is all about integration between multiple sensory inputs and can never be achieved by any one alone. That was behind Tesla’s initial avoidance of Lidar on the basis that its rivals appeared to be over dependent on the technology.
That was the red herring, because the point was that any autonomous diving venture over-reliant on Lidar was doomed and that the same would apply to any other sensory input. In fact, Tesla has now admitted Lidar to its tool set on that very basis, acknowledging that the more sensory inputs there are, the merrier.
The actual challenge is effectively to integrate these multiple sources to generate that all important dynamic 3D view of the surrounding environment as the foundation of ambitions to make autonomous driving far safer. Human error arises partly from incomplete visibility and autonomous vehicles should in principle be able to eliminate that, as well as being able to judge distances and speeds more finely. But human drivers, especially experienced ones, are better placed in some cases to deal with unexpected situations arising and it will be years before autonomous vehicles can close that gap entirely.
There is also radar, using radio signals to locate objects and distances, and sonar generating sound waves to do the same. They all have pros and cons but given the highly safety critical aspect of self-driving, the most effective systems will incorporate all of these. If one fails in a given situation a sufficiently accurate and dynamic 3D representation can be generated with the others. That is why comparisons presenting these technologies as alternatives are wide of the mark and it their complementary properties that are relevant.
For example, Lidar is very good at detecting small objects because of the short wavelengths of the light and also has high precision, making it good at generating accurate 3D monochromatic images of the target. Radar on the other hand can operate under more challenging conditions such as fog and can deal with dirt temporarily clogging a sensor. Sonar has the advantage over light based systems that it is not confused by transparent or reflective objects, but has far higher wavelength at around 4mm instead of 550nm for visible light, so that it is only reflected accurately by rougher surfaces.
The overall integrated imaging system has to cope with varying light and artificial illumination of the surroundings under darkness. It should also have vision in conditions of poor visibility and beyond line of sight, so that it can go further than human drivers in being able to obtain advance warning of hazards not yet in direct view, around bends in the road for example. The latter is where the 5G C-V2X technology will play an important role, alongside conventional optical cameras complementing Lidar.
Lidar though has a big role to play because of its high precision and it is notable that Volvo considers it essential to progress towards Level 3 driving where autonomy is achieved under some circumstances but a human still has to be at the wheel ready to take over. It is notable that Audi has already incorporated lidar sensors called Scala, but these are from an earlier generation with poorer resolution and maximum range of around 80 meters. While such basic Lidar systems like this enable some driver-assistance features including emergency braking, Volvo considers them good enough only for Level 2 driving where autonomous steering, acceleration and braking can be performed, but the driver must be at the wheel, fully alert and in full control.
Level 3 allows drivers to switch off and read a book if they want, but still be ready to take back control at short notice. Level 3 then is the real transitional phase and problematic because there is a danger of drivers nodding off and not being ready to take over quickly enough if an unforeseen situation arises that needs rapid action.
With its more advanced Lidar, Volvo’s cars will be ready for Level 3, which also implies geofencing such that autonomous driving is restricted to certain roads and will not be allowed everywhere. Indications are that Volvo’s cars will be on a similar page as, say, GM’s super Cruise technology, which is limited to roads pre-mapped by the manufacturer, so that operation would be linked presumably to the satellite navigation system.
The road beyond that to Level 4 where drivers can hand over full control under most circumstances and then Level 5 where manual controls such as steering wheels and brakes are removed will be long, with many obstacles to overcome.