Elon Musk dropped some interesting phrases at a recent event for Tesla investors, among which was the revelation that Lidar, a laser-based scanning technology that creates 3D images of objects, is "something stupid" and that "anyone who trusts LiDAR is doomed. " This sounded quite radical given the number of autonomous cars that rely on technology, but researchers from Cornell University, United States, have endorsed Musk's predictions thanks to a new method for autonomous cars to see the world in 3D using a couple of cheap cameras.
A few autonomous cars are enough to dramatically reduce traffic jams
Human beings are united by our infinite capacity to do the opposite of what they order …
Read more Read
Being able to visualize and detect objects around a vehicle in three dimensions is crucial for autonomous cars to operate safely, in a world where roads are shared with other vehicles, cyclists and, often, pedestrians. As a driver, every time you turn your head to scan what is around your car, your brain instantly visualizes your surroundings in 3D and assesses the possible dangers. The use of cheap sensors to detect objects near an autonomous car is not enough. When you travel on the road at 100 kilometers per hour, the car must see what is approaching and thus be able to plan what is necessary to avoid the dangers.
That is why it is common to see Lidar systems (light and range detection systems) placed on autonomous cars. Using rotating lasers, they scan the surroundings of a vehicle and generate 3D images of near and far objects, allowing the software to analyze the results and identify things that should be avoided. However, Lidar is very expensive, often adding about $ 10,000 in additional parts to the price of a car, and needs to be positioned on the vehicle to get the best perspective. Currently the industry is trying to maximize the range of electric cars and gas, so a Lidar system adds a lot of resistance to the aerodynamics of a car and its performance.
In a study to be presented in June at the conference on computer vision and pattern recognition in 2019, called Pseudo-LiDAR of visual depth estimation: reducing the detection gap of 3D objects for autonomous driving, Cornell researchers detail a possible breakthrough for autonomous cars. Cameras have generally been considered a technology inferior to Lidar, since they are often installed at low angles near the car's bumper, which results in images that tend to distort objects in the distance, something that confuses networks neurons that try to process and interpret the data.
This is what happens when a self-driving car tries to give way to the highway
The following images will break your heart. An autonomous Waymo vehicle tries to get up …
Read more Read
But by placing a pair of cameras on either side of a vehicle behind your windshield, stereoscopic images are produced that can be converted to 3D data. Because the images are generated from a higher point of view, closer to where the Lidar systems are usually installed, it was determined that the 3D data, which were generated from the cameras, were almost as accurate as what the laser scanners can generate, without distortion, and with a much lower cost.
However, it will probably be a long time before this research reaches the autonomous cars. Lidar remains reliable and incredibly accurate, and the companies that work in these cars are currently more concerned with safety than with costs. But as technology and software improve, and the restrictions that limit where and when autonomous cars can circulate are disappearing, driving autonomy will soon be a major concern for consumers looking to buy a new car, and they Yes, they worry about the costs. The Cornell researchers' approach will make it much cheaper to implement autonomy features in cars, and perhaps over time could make Lidar obsolete.
Maybe Musk was right?