Tesla has released update 10.11 for its Full Self-Driving (FSD) self-driving system.
According to Andrey Karpaty, director of the company for artificial intelligence, one of the significant changes in FSD was the modernization of lane geometry modeling from dense rasters (a set of points) to an autoregressive decoder. Thanks to this, the system directly predicts and connects the strips of the “vector space” point by point using a neural network-transformer.
Other changes in the update include a reduction in the number of false car decelerations, improved lane understanding when the map is inaccurate, and an increase in the accuracy of recognition of pedestrians and cyclists. The smoothness of turns is also optimized.
Recall that in January, California authorities announced their intention to ban Tesla from testing an autopilot in the state without permission.
In September 2021, scientists found that drivers with FSD turned on are more likely to lose their vigilance on the road.
In the same month, an unmanned Tesla drove through the center of Kyiv thanks to an autopilot leaked into the network.