Sensors & Sensibilities
Driverless cars need all the sensors they can get — cost and style don’t matter (yet)
Self driving cars are not beautiful. I mean that purely aesthetically. Creating safe self driving cars will be a beautiful achievement, but the plethora of sensors protruding from otherwise sleek vehicles rankles with our expectations of smooth, clean car designs. From an engineering perspective, though, they are very beautiful — as part of one of the most advanced fields of development challenging human ingenuity to, and sometimes beyond, its maximum capacity.
Sensor Fusion
Sensors are key to self driving cars. Without some combination of radar, sonar, lidar (Light Detection and Ranging — a sensor that uses a pulsed laser to measure distance and detect objects) and cameras, they can’t see where they are and what’s happening around them. Without those crucial inputs, they can’t decide how to proceed. But with them, using complex AI techniques, engineers can create a pretty good perception snapshot using what’s known as Sensor Fusion.
With most prototypes opting for lidar units positioned on the roof to get a clear field of view around the car, these appendages create vehicles that are instantly recognisable as robocars. This could be a good and bad thing. If it’s obviously a robocar, humans can adjust their behaviour around it, knowing that the robots are still like learner drivers. Those with a grudge against the robots however, can also easily spot their quarry.
In Camera
“Lidar is a fool’s errand. Anyone relying on lidar is doomed. Doomed! [They are] expensive sensors that are unnecessary. It’s like having a whole bunch of expensive appendices… it’s ridiculous, you’ll see.”
Elon Musk
Tesla has been singular in its opposition to the use of lidar, believing that cameras can provide the required inputs. You can only see the cameras on a Tesla if you go looking for them. They are beautifully integrated, barely visible in the door pillars. Although dismissed ostensibly on the grounds of cost, I can’t help but wonder how much of the opposition is due to design sensibilities. The cost argument doesn’t really stack up. It did when lidar units cost more than the vehicles themselves — just a few years ago a lidar would set you back some $70,000, and you typically needed several of them per car. With lidar now less than 10% of that cost and trending towards sub-$1000, adding them to the sensor array is clearly viable.
Cost is Irrelevant
For me, the cost argument against lidar is irrelevant anyway. While, clearly, we need to have a path to commercial viability, creating the safest vehicles practical matters more than creating the cheapest possible cars. We already add significantly to the cost of conventional cars in the name of safety. We know that costs will fall over time for any technology. But we can’t start cutting corners and optimising producibility until we have the basic technology working reliably.
Given that the CDC estimates the financial costs of US road traffic deaths at over $44bn annually, less than a week at that rate would fund Waymo’s R&D budget for a year. That would reduce to a few hours’ costs if you include the financial cost of injuries as well as fatalities.
Or to look at it another way, if the equivalent of the lifetime cost of insurance each driver pays were to be applied to the prevention of crashes, rather than paying for their aftermath, you could easily justify even the most costly of sensors. So, even setting aside the human cost of road deaths and injuries, penny-pinching on the cost of technology with the potential to reduce crashes substantially is morally questionable.
Crutch or Stabiliser?
Tesla has likened Lidar to a “Crutch”. But I don’t think a crutch is necessarily a bad thing. When you’re unsteady, unable to walk on your own, then of course you use a crutch. You work to develop your abilities to the point where you don’t need it anymore, but dismissing a crutch and thinking you can run before you can walk — that’s the foolish option. I liken it to a stabilizer when learning to ride a bike — your ultimate ambition is, of course, to remove the unsightly stabilizers. But the risk of injury is too great to take them off before you’ve mastered the basics.
Multiple levels of safety are surely preferable, even if they may be redundant — we sometimes add rumble strips to painted road lines to add another layer of feedback. Failing to use lidar as well as cameras and radar seems to me to be akin to saying “we already have lines; rumble strips cost more and aren’t very elegant”. Someday we probably will be able to create driverless cars that rely primarily on cameras to see. We’ll solve the perception, depth and other shortcomings they currently face. But as lidar prices continue to fall, and performance continues to improve, it’s hard to argue that we shouldn’t gratefully use the crutch.
“Lidar won’t be needed in five years, given how good cameras and radar will get. But for now we will continue to use lidar. We’re going to develop lidar until we don’t need to.”
Eric Meyhofer, Head of Uber’s Advanced Technology Group
Effectiveness Trumps Elegance
As is often the case, our ambition is ahead of current technology. For now though, it’s about effectiveness, not elegance. Companies developing self driving cars need to overcompensate on safety before they focus on cost and appearance. This may be hard at a time when they are also craving public acceptance from a sceptical audience who don’t want strange looking appendages on their cars. But I’d much prefer the dome of a lidar or camera array over my head than the dome of an ambulance emergency light.
