The first fatal accident of a Tesla Model S in Autopilot mode triggered a hail of criticism on the system created by the constructor. Unlike many systems such as those created by Google or automotive suppliers, this system is essentially based on cameras and radar. The absence of laser sensor, the costly Lidar. The Autopilot is not infallible and researchers are even fun to lure it. The Tesla system is certainly still immature but, despite expectations of the public, the algorithms will not prevent all accidents.
The Tesla Model S sensors failed to detect the fatal obstacle
The absence of a LIDAR (an acronym for “laser detection and ranging” or laser sensor) on the Tesla Model S is responsible for the death of Joshua Brown? It is the assumption made by many experts who point out that the sensors available to the rugged Model S could not detect the trailer of the truck that crossed its path. It, too high to be detected by radar, could be spotted by the front camera of the car due to a strong backlight, so the car continued its path, causing its occupant in death.
In these circumstances, a LIDAR would have definitely helped the car to identify the deadly obstacle and caused an avoidance manoeuvre or stop of the vehicle. By a macabre coincidence, a Tesla surmounted by a LIDAR was reported a few days later in Palo Alto, not far from the headquarters of Tesla Motors, evidence that the manufacturer was already working effectively on the installation of one or more LIDAR on its future cars.
This tragic accident would have also led Elon Musk to terminate its partnership with Mobieye, the company that provides image analysis algorithms that equips the current Tesla. A blow to the Israeli company, also partner of the french Valeo, and which was thus indirectly pointed to by the founder of Tesla Motors as responsible for the accident.
Interference of the sensors accidental or voluntary, a real risk
In addition to the limitations of each sensor, researchers from the universities of South Carolina and Zhejiang in China showed that it is relatively easy to lure a Tesla Model S sensors. Their study, unveiled at the Black Hat conference, dismounts with adequate facilities have can literally do appear in the eyes of a fictitious barriers Tesla on a road or even conceal real obstacles. Small pranksters can, with a laptop computer and $40 equipment induce automatic parking of the car in error and prevent it from Park at a location yet wide enough to accommodate the car. Worse, with more complex radio equipment it is possible to erase other vehicles on the road, with the consequences that we imagine. This interference of the sensors can potentially be used for other types of vehicles.
Elon Musk promises a total automation of vehicles in the short term
To counter these threats and avoid than an accident such as that known driver misfortune of the Tesla, manufacturers will have to multiply still different and complementary sensors: optical, laser, radar and ultrasonic. They will also be able to feed their algorithms with data generated by the infrastructure and other vehicles, but then still new problems will arise. With computer and mobile, the public is accustomed to the fast evolution of technology but with the autonomous cars it will take many years before industrialists hold the promise of a robot car where you can really do without a steering wheel. Experts estimate that fully autonomous cars will appear only in 2025. A roadmap that Elon Musk wants to shake up considering that the time of autonomous cars of level 4, it means completely robotic, nearby.
Translation : Bing Translator
“Elon Musk Hints That Tesla Updates Will Soon Lead To Level 4 Autonomy”, Futurism, August 7, 2016
“Hackers Fool Tesla S’s Autopilot to Hide and Spoof Obstacles”, Wired, Augsut 4, 2016
“Tesla Divorces Mobileye, or Vice Versa”, IEEESpectrum, July 26, 2016
“A Tesla Model S with Lidar spotted on the road around Palo Alto”, TechCrunch, July 7, 2016
“Tesla Autopilot Fatality Shows Why Lidar And V2V Will Be Necessary For Autonomous Cars”, Forbes, July 1st, 2016