Controlling self-driving cars with cameras.. the “Tesla” option raises the suspicion...

Tesla is betting on cameras to control future self-driving cars, raising the suspicion of some specialists in control systems that rely on radar and remote sensing technology.
Luminar, a manufacturer of lidar (laser sensing) devices, conducted a live experiment at the CES Las Vegas Electronics Show to prove the superiority of its technology, launching two cars into a parking lot at a speed of about 50 kilometers per hour before displaying a doll in the form of a child in front of them.
The lidar-equipped car stopped in time, while the other Tesla car hit the dummy.
An expert outside the company did not verify the conditions of the “Luminar” experiment.
“We didn’t want to show the experience via PowerPoint or in a nice video,” Aaron Jefferson, who is responsible for product development at Luminar, told AFP.
He adds that “the cameras can be very effective when driving in ideal conditions during a sunny day,” considering that “the problem lies in the unusual situations” facing the car, which are unclear turns, fog, plastic bags, light at sunset, and others. .
Most manufacturers of autonomous systems have chosen to combine cameras powered by radar and/or lidar, which are instruments that allow distance measurements via radio waves or lasers.
Last year, Tesla abandoned radars and relied solely on cameras in its ride control system.
Elon Musk considers that with technological progress, an “artificial brain” that works with cameras can match the capabilities of the human brain, which analyzes its surroundings through its eyes.
“This strategy makes a lot of sense,” says Cornell University professor Kellyanne Weinberger, who has worked on object sensing in autonomous driving systems.
Officially, Tesla currently only offers ride control systems, but it hopes to reach a fully autonomous driving system.
Several years ago, Tesla chose to automatically install cameras and radars on all of its cars, and thus was able to obtain a great deal of information about the way drivers drive their cars in real-world conditions.
Weinberger says: “Tesla bet that by collecting a lot of data, the cameras can create an algorithm as efficient as the algorithm that uses more expensive sensors, but with less data.”
For example, Waymo’s self-driving car company, Google’s self-driving car company, is packed with sensors, but only works under certain conditions.
Sam Abu Samad of Guidehouse Insights points out that autonomous driving systems have four main functions: perceiving the environment, predicting what will happen, planning what the car will do, and executing.
“The prediction turns out to be more complicated than the engineers thought, especially with pedestrians and cyclists,” he says.
The advances that engineers thought they could make in programs that run only on cameras have stabilized at some level through artificial intelligence and machine learning.
Abu Samad considers that the problem is that “Elon Musk showed his company’s driving system in a way that shows that the equipment installed in his cars will be sufficient,” adding that “Tesla cannot back down, because hundreds of thousands of people paid money” to get it.
“Cameras alone are not enough, no matter how much data is stored in them,” said the head of the French equipment manufacturer Valeo, which displayed its third generation of lidar at the Las Vegas Electronics Show.
“Understanding what is happening around the car, what is seen and not seen, and analyzing it all day and night, is very essential,” Jack Aschenbrois told AFP.
“It is our absolute conviction that lidar is needed” to achieve more advanced levels of autonomy, he adds.
“All sensors have their advantages and disadvantages,” said Marco Bertona, a professor at Italy’s Unimore University and head of a team running an unmanned vehicle that competed in a self-driving car race in Las Vegas on Friday.
He says that cameras “in the current prevailing knowledge” alone make many mistakes.
Currently, he adds, “the more systems you have running in parallel, the more you are able to integrate different types of sensors, and therefore the more likely you are to be among the first to adopt the safety requirements that regulators will impose.”