2022.20.8 Tesla Official Release Notes
– Improved decision making for unprotected left turns by using better estimation of ego’s interaction with other objects throughout the maneuver.
– Improved stop pose while yielding to pass through objects during unprotected left turns “à la Chuck Cook” using the middle safe zones.
– Made the speed profile more comfortable when crawling for visibility, to allow smoother stops when protecting potentially occluded objects.
– Enabled crawling for visibility at any intersection where objects might cross ego’s path, regardless of the presence of traffic controls.
– Improved lane position error by 5% and lane recall by 12% with a complete vector lane neural network update. Information bottlenecks in the network architecture have been reduced by increasing the size of per-camera feature extractors, video modules, and GPT attention module internals.
– Improved crossing and merging lane position error by 22% by adding long range jump connections and a more powerful trunk to the network architecture.
– Improved the speed error of pedestrians and cyclists by 17%, especially when the ego makes a turn, by improving the on-board trajectory estimation used as input to the neural network.
– Improved animal detection recall by 34% and decreased false positives by 8% by doubling the size of the self-labeled training set.
– Improved detection recall of vehicles crossing away from 4% by adjusting the loss function used during training and improving the quality of the tag.
– Improved the “is parked” attribute for vehicles by 5% by adding 20% more examples to the training set.
– Upgraded occupancy network to detect dynamic objects and improve performance by adding a video module, adjusting the loss function and adding 37,000 new clips to the training set.
– Reduced false slowdowns around crosswalks due to better classification of pedestrians and cyclists as not intending to interact with ego.
– Reduced false lane changes for cones or jams by preferring a slight shift in lane when applicable.
– Improved lane positioning on wide residential roads.
– Improved object future path prediction in scenarios with high yaw rate.
– Improved accuracy of speed limit signs on digital speed limits by 29%, hard relevance signs by 23%, 3-digit speeds by 39%, and end of speed limit signs 62% speed. The neural network was trained with 84% more examples in the training set and with architectural changes that allocated more computation in the head end.
Tap the “Video Record” button on the top bar UI to share your feedback. When pressed, your vehicle’s external cameras share a short snapshot of the Autopilot associated with the VIN with Tesla’s engineering team to help make improvements to the FSD. You will not be able to view the clip.