Tesla Begins Rolling Out Fsd Beta 10.13

Tesla has begun rolling out version 10.13 of its highly anticipated Advanced Driver Assistance System FSD Beta and most of the updated release notes have been shared online. And Whole Mars, a Tesla FSD Beta tester, has shared the release notes for Tesla FSD Beta 10.13 on Twitter.

According to Tesla CEO Elon Musk, the release notes for FSD Beta 10.13 directly reference Beta tester Chuck Cook's challenging unprotected left turns, as well as other notable improvements regarding false alarms and false decelerations around crosswalks. While FSD Beta v10.13 will by no means be the final version of this advanced driver assistance system, it promises to make significant improvements overall.

The following are some of the release notes for FSD Beta v10.13.

Improved unprotected left turn decision making by maneuvering to better evaluate vehicle interactions with other objects.

Improved stopping position during "Chuck Cook style" unprotected left turns by utilizing intermediate safety zones.

Creeping to improve visibility makes the speed profile more comfortable for a smoother stop when protecting a potentially obscured target.

Enable creeping to improve visibility at intersections where targets may cross vehicle paths, regardless of the presence of traffic control.

Improve lane position error by 5% and lane recall by 12%.

Improved lane position error at intersections and merging lanes by 22% by adding remote skip connections and a more robust backbone to the network architecture.

Improved speed errors for pedestrians and cyclists by 17% by improving the in-vehicle trajectory estimation used as input to the neural network, especially during vehicle turns.

By doubling the size of the automatic marker training set, animal detection recall improved by 34% and false alarm rate was reduced by 8%.

By adjusting the loss function used during training and improving tag quality, the detection recall of distant crossover vehicles was improved by 4%.

Improved the "parked" attribute of vehicles by 5% by adding 20% more examples to the training set.

Upgraded the occupancy network to detect dynamic objects and improve performance by adding a video module to the training set, adjusting the loss function, and adding 37k new segments.

Reduced false slowdowns near crosswalks by better classifying pedestrians and cyclists as objects that do not interact with vehicles.

Reduced false lane changes from cones or obstructions by gently offsetting within lanes where appropriate.

Improved in-lane positioning on wide residential roads.

Improved future path prediction for targets in high yaw rate scenarios.

Improved the recognition accuracy of numeric speed limit signs by 29%, non-correlated speed limit signs by 23%, three-digit speed limit signs by 39%, and end of speed limit signs by 62%. The neural network was trained using an additional 84% of the examples in the training set, and the architecture was changed to assign more computations to the head of the network.

Discover our Range


View all


View all