Tesla has announced that Model 3 and Y produced from October 2022 will no longer be equipped with either radar or ultrasonic distance sensors. They are to be replaced with a system based exclusively on cameras, the only ones in the world.
Tesla chooses its own path
Elon Musk has stated that a camera-based surveillance system can do as good a job as the human eye, while condemning LIDAR - the system that all other car manufacturers rely on. Tesla has so far used ultrasonic distance sensors together with cameras. Since the sensors are mainly used for parking assistance, the loss should not result in any change to the safety level. Unfortunately, Tesla customers with brand new cars will not have access to the parking alarm, self-parking or summon functions for the time being. This will be supplemented in later software updates.
Musk claims that the camera-based system, Tesla Vision, will be far less expensive than laser and radar-based systems – and make cars cheaper for most people. He predicts that LIDAR will lose steam Tesla Vision and mentions several reasons. Among other things, camera-based systems will provide better accuracy, for example being able to distinguish between a plastic bag and a bump in the road, and more easily see which way the object is moving. Furthermore, camera-based systems will be more precise and dynamic than LIDAR, which is based, among other things, on static map data.
Unjustified sudden stops
In several forums it is now being discussed whether new Teslas are as safe as before. It is reported that several cars without sensors have made unmotivated sudden stops - so-called phantom braking. One wonders whether all previous safety functions will continue to function, for example warning for oncoming traffic when turning left at an intersection.
TO DEAL
LIDAR-based systems are based on a dense belt of laser signals where the return signals indicate the distance between the objects. The density of the signal flow makes it possible to draw the contours of each object, but gives little information about what lies within the contours. LIDAR's strength is that it provides reliable distance measurements and size of the objects - and can tell with a high degree of certainty whether you are on a collision course with something. RADAR follows the same principle as LIDAR, but transmits radio signals instead of laser. The radio signals can reach greater distances, but give a less accurate picture of the objects.
Tesla Vision
Vision-based systems are based on two-dimensional images of the same thing that we see with our own eyes, actually better than the human eye can, including in poor lighting conditions. The disadvantage is that the solution does not provide any exact distance measurements. It makes high demands on software and computing power. Image processing is demanding, and there is a risk that the processing takes a disproportionately long time when many, rapid and large changes occur at the same time, which is predictable in critical situations - when it matters most.
One such case is when someone walks out into the road just after you have turned the car into a side street.
Then everyone missed
In 2019, AAA in the USA tested several car models that all failed in this situation. They let a doll cross the road just as the test car turned the corner. None of the cars in the test were programmed to handle this situation. Car manufacturers understandably prioritize the same tests as Euro NCAP and IIHS because it is de which assesses the car models' safety level. Euro NCAP tests various scenarios with collisions from the rear between vehicles. The Insurance Institute for Highway Safety (IIHS) covers more scenarios than Euro NCAP, including left turns into crossing traffic and incidents between cars and pedestrians. It is especially the last that seems to cause the most problems.
AAA tested these scenarios in 2019
American Automobile Association (AAA) is the American equivalent of NAF. The tests they did in 2019 were of the same type as the IIHS performs, but not exactly identical. They show that the cars are not (were) as safe as the results in Euro NCAP and IIHS indicate. The AAA tests included the following:
Children who are thrown between two street-parked cars while the car is traveling at 32 km/h. 90% failed.
Children crossing the road at the same time as a car turns around a street corner at 32 km/h. 100% failed.
Two pedestrians walk next to each other while a car approaches from behind at 32 km/h. 80% failed. The same test gave even worse results in the dark.
In 2019, the Tesla Model 3 did neither better nor worse than the other models.
Both?
The problem with Tesla Vision is that the system should interpret 2-dimensional images into 3D. It often goes wrong, as for example when I recently drove on a Class B motorway where the white border strip had a thin strip on the outside - perhaps from a mis-marking. The Tesla interpreted this as a high edge and marked it on the screen. Every time I approached the curb at 100 km/h the collision warning screeched. LIDAR would understand that this was no physical obstacle. LIDAR, for its part, can alert the driver when leaves or light plastic are blowing over the road. Tesla Vision supported with LIDAR would be an ideal solution – at least on more expensive Tesla models. But it is unlikely to work since the company manager has ruled LIDAR north and down?
Do not try this at home!
Testing automatic emergency stops is not something you should do at home. Leave it to the pros. I must admit, however, that I have seen emergency stops a couple of times in connection with pedestrian crossings where there is a close distance between pedestrians and cars. The situations have arisen when I have walked in a row towards the pedestrian crossing with dogs and stopped to let the cars drive over first - both times with eye contact with those behind the wheel. I don't know if it's my brisk walk or the dogs' movements, but the cars chose to oversteer the drivers in such a way that the people sitting in the cars were hanging in their seat belts, terrified. A special feature of the emergency stop is that the wheels are not locked after the sudden stop before the driver puts his foot on the brake. In the shock, it usually takes a couple of seconds. Only then can you go over.
Facts IIHS
The IIHS (Insurance Institute of Highway Security) includes the following tests between vehicles and pedestrians.
IIHS tests in 2019 and 2022
Tesla Model 3
In 2019, the car failed quite badly when crossing children. Only a minimal speed reduction was recorded in both 20 and 40 km/h. In the case of an adult pedestrian in the same lane, the car slowed down so much from 60 km/h that it only swerved into the pedestrians. In 2022, Tesla passed all tests without hitting any of the pedestrians.
Volvo XC40 (LIDAR)
The tests in 2019 and 2022 gave exactly the same results. The Volvo passed all the tests, but ran into pedestrians on two occasions.
Hyundai Kona (LIDAR)
The results for 2019 and 2022 are almost identical. Hyundai passed tests carried out at low speed (20 km/h). At 40 km/h, the car was unable to slow down sufficiently and hit the victims at 11 km/h. With an adult pedestrian parallel in the same lane, the car slowed down from 60 km/h to 37 km/h before it ran into the victim. In tests like this, the cars get points for detecting the pedestrian and managing to reduce their speed. Hyundai scored worse than Tesla and Volvo, but was still rated "Advanced".
See
CBS News from October 2019