Musk: Tesla has started to push FSD Beta V11, bringing a lot of feature updates and optimizations.

According to CTOnews.com news on November 13, Tesla has begun to push the FSD Beta V11 version of the driving assistance system to testers, and overseas users will be able to receive pushes one after another since 15:11 on the 12th. Musk also confirmed that Tesla will be "widely launched" to more North American users in a few weeks.

For the FSD V11 version, Musk previously pointed out on the 2022 AI that the V11 will use a single-stack system, which will make FSD Beta performance smoother and smarter judgment. Therefore, most of the industry believes that FSD Beta V11 will be a major update of Tesla's advanced driving assistance system.

CTOnews.com learned that Musk said at the time that he was using a single-stack version of FSD Beta, so he was able to reach his destination without touching the vehicle control.

According to the release notes, FSD Beta V11 (2022.40.5) has been updated and optimized mainly in the following areas:

Enable FSD Beta on the highway.

This unifies the vision and planning stack on and off the highway and replaces the four-year-old legacy highway stack. Traditional highway stacks still rely on several single cameras and single-frame networks and are set to handle simple lane-specific operations. FSD Beta's multi-camera video network and next-generation planning system allow more complex agent interactions with less lane dependence, making way for smarter behavior, smoother control, and better decision-making.

The accuracy of the space occupying network (Occupancy Network) in the recall of close obstacles and severe weather conditions has been improved, the spatial resolution of transformer has been increased by 4 times, the capacity of image features has been increased by 20%, the calibration of side cameras has been improved, and 260000 video training clips (including actual and simulated) have been added.

The prediction speed errors of very close-up motorcycles, scooters, wheelchairs and pedestrians were reduced by 63.6%. For this reason, we introduce a new data set that simulates adversarial high-speed VRU interactions. This update improves the autopilot control system around fast moving and cutting VRU.

Improved crawl configuration (creeping profile) to increase acceleration at the beginning and end.

By using the general static obstacle network to predict the continuous distance from the static geometry, the control of nearby obstacles is improved.

By increasing the dataset size by 14%, the error rate of vehicle "parking" attribute is reduced by 17%, and the accuracy of brake lights is also improved.

The performance of the difficult scene is improved by adjusting the loss function, the speed error of the clear scene is increased by 5%, and the speed error of the highway scene is increased by 10%.

The detection and control of opening doors are improved.

In the calculation of vehicle kinematics, the optimization-based method is used to determine which routes have nothing to do with calculating the given transverse and longitudinal acceleration and steep rise limits, so as to improve the stability of the turn.

The Ethernet data transmission pipeline is optimized by 15%, thus improving the stability of FSD UI visualization.

The recall rate of the vehicle behind the ego is improved, and the accuracy of the vehicle detection network is improved.