During a November 10th Twitter session, representatives for Waymo responded to the question, “Would you say that your technology is orders of magnitude more advanced than the more vocal competitor with a misleading branding?” with a simple, “Yes.”
Waymo is a subsidiary of Alphabet, the parent company of Google, and its working to develop self-driving technology through ongoing advances in its fleet of taxis in Arizona. The more vocal competitor the questioner was referring to is Tesla, who recently released a beta version of its Full Self-Driving (FSD) system to a select group of people driving its cars. And it’s that FSD the questioner was probably alluding to as misleading branding.
The main difference between Waymo and Tesla’s technologies is that Waymo relies heavily on lidar to keep its vehicles safely within their lanes and to detect objects they need to avoid, while Tesla’s vehicles use a combination of cameras and radar. Elon Musk, Tesla’s CEO, has gone on record against lidar, suggesting that cameras are the only feasible route forward for autonomous vehicles. But many experts in the field insist lidar’s shortcomings are well on their way to being overcome, while the issues with cameras are here to stay.
Lidar in Space, Cameras on the Road
Musk, despite his outspoken skepticism of lidar in self-driving cars, knows the technology has its applications. SpaceX, his other company, equips its shuttles with lidar that helps them dock with the International Space Station. Why are these sensors good enough for spacecraft but not for navigating through traffic here on Earth? For one, docking a spaceship is a slow process that requires a great deal of precision. Driving a car demands quicker responses and is usually more forgiving of minor imprecisions.
Plus, in space, you seldom have to worry about weather conditions. Waymo chose Arizona for their deployment of self-driving taxis for a reason. Its vehicles can now handle a little bit of light rain, but too many droplets in the air and the point clouds generated by the lidar’s software become unreadable. This is because lidar works by emitting light beams and tracking the time to each beam’s reflection. When the light beams bounce off droplets in the air, it throws off the estimates of distance. (To be fair, rain and snow can trip up camera systems too, just not as severely.)
Level 4 in Arizona or Level 3 Everywhere
Still, on the surface, it looks like Waymo’s self-driving capabilities are superior to Tesla’s. Waymo’s vehicles have achieved Level 4 autonomy, meaning the vehicles can handle themselves in almost all circumstances, but they still need drivers onboard to take over if they encounter some edge case they’re not programmed for. (This is already starting to change.) Tesla’s Autopilot, meanwhile, and even its FSD, is probably at Level 3, meaning the human driver is supposed to keep alert, usually with his or her hands on the wheel.
But there’s another difference. Tesla has cars on the road across the country—and beyond. This is because while Waymo’s technology relies heavily on detailed maps created prior to each drive, Tesla is trying to achieve far greater flexibility by taking something of a brute force approach with its machine learning programs. The self-driving systems in Tesla’s vehicles are dependent on training from countless hours of data collection on the road. Waymo meanwhile matches its lidar data to those premade maps, thus achieving more reliable results but with much less flexibility.
The deeper source of this difference is that both companies are using the resources available to them. Waymo has access to all those maps Google’s vehicles create on the streets. Tesla has countless vehicles already on the road recording data that can be used to train its driving programs.
But which of these two approaches will pay dividends?
Lidar vs. Cameras
Lidar provides a more direct measure of distance to objects, and this means quicker and more reliable responses on the road. This is the biggest reason so many experts think lidar is indispensable for self-driving vehicles. Cameras rely on at least two vantage points and software that calculates parallax to determine distance. While there is research suggesting this approach can be just as effective, it still demands more processing, which can lead to more complications. Some machine vision companies are using entire camera arrays instead of just two, but how much potential any of these technologies hold is yet to be seen.
Meanwhile, lidar still has other problems. The main one up till now has been cost. It takes around $10,000 to equip a vehicle with a lidar system. Analysts point out this is almost sure to change as the technology advances, but for now cameras are far less expensive. Another problem with lidar is that it takes a lot of energy to power. Many of the autonomous vehicle makers are also competing to make the most efficient electronic vehicles at the same time, so a system that draws huge amounts of power threatens to reduce how far these vehicles can travel between chargings.
There’s also the problem of street signs and traffic lights. Lidar can detect the shape of signs, but it can’t discern one message from another. And it can’t tell red from green.
Cameras have issues of their own. While their performance at determining distance is close to that of lidar, every little bit counts when it comes to road safety. Small differences in outcome between the two technologies can result in vastly different numbers of collisions at scale. Cameras may also be somewhat better when it comes to rain and snow, but that doesn’t mean they’re great in these conditions. (Nor, for that matter, are human visual systems.)
The biggest remaining issue for cameras is that no one knows if machine learning can ultimately deliver on its promise to turn 2-D images into 3-D models of the environment accurately enough to consistently avoid collisions. Musk is betting that it can and that it’s only a matter of time before it does. That’s why he has all the Teslas on the road today recording data to train the systems.
So, which approach is favored to win the race to full autonomy? Most companies are betting on lidar, but a lot of analysts still believe it’s unwise to count out Tesla. (Another AV company using cameras is MobilEye.) And the winner may not end up being a company that uses one or the other. The next stage in the march toward self-driving vehicles may focus on combining data from multiple types of sensors.
Also Check Out:
And
And
Autonomous Vehicle Tradeoffs
Follow us on:
Twitter: https://twitter.com/ConvoyTechnolo1
You can also sign up for our newsletter, so you never miss a post. Just scroll down to the form in the footer.