
The American company Tesla Motors has released a new version of the software for its electric vehicles - Version 8, which has implemented a new on -board radar. As explained in the company’s blog, the autopilot algorithm is now relied on to the radar as one of the main sources of information about the road and objects on it. In addition, an algorithm of group training of cars with memorizing safe, suspicious and dangerous places has been introduced, N+1 reports.
Since 2015, when the first version of the autopilot was released , the main source of information about the Tesla cars was a term paper, and the radar station (radar) was only auxiliary system. This was explained by the fact that "the world was too strange in radar": for example, people became partially transparent, and objects made of wood and plastic were classified as glass.
The course chamber helps to avoid these errors, adjusting the image. But she becomes useless in snow, fog or rain, while the radar continues to work stably.
Tesla Motors stated that they solved this problem by increasing the frequency of information: now data from the radar come every 100 milliseconds - six times more often than before. In addition, from the shots from the radar, which come every tenth of a fraction of a second, 3D-facing in the form of a mobile cloud of points are folded, which will distinguish moving objects from stationary, as well as recognize false reflections.
As you know, in May of this year, the first accident occurred in Florida with the participation of the Tesla Model S sedan, in which the driver of the electric car died. Tesla in the autopilot mode entered under the semi -trailer a tractor that appeared in front of her at the intersection. According to preliminary data from the investigation, the electric vehicle camera took the semi -trailer painted in white. In turn, the radar of the autopilot system “did not see” the semi -trailer, since its elevation angle is specially underestimated so that the machine does not react to braking to various low objects.
In order to avoid the repetition of such accidents in the future and the false response of the autopilot to the overlapping of tunnels, Tesla Motors introduced group training. The first time after the update for all Tesla cars will not be responded to potentially dangerous ones from the point of view of the system system, such as entrances to the tunnels. If Tesla drivers, passing such places several times, will slow down, all electric cars, who have never been in these places before, will also slow down in the autopilot mode. All this information will constantly analyze the Tesla onboard computer and if in some new situation it considers the probability of a collision very high, the brakes will be automatically involved.
According to Tesla Motors, this will not allow you to completely avoid accidents, but it will help reduce the speed before the accident to a safe level. In general, Tesla software refinement, including group training and the active use of the radar, as the company promises, will lead to the fact that the autopilot will learn to use the brakes correctly, "even if a UFO in a thick fog sits in front of the machine."