Tesla owners test FSD Beta 10.69.25 during the cold Christmas holidays (release notes, videos)

Tesla gave a gift to its FSD Beta testers on Christmas by rolling out a big improvement software update version 10.69.25 (firmware version: 2022.44.25.5). Since Tesla and Elon Musk love easter eggs, 25 in this update version number refers to the Christmas date.

Tesla has also provided a long list of release notes that detail specifics of each and every improvement the automaker has made in FSD Beta 10.69.25 (read the release notes below).

As the weather during this end-of-year Holiday Season has become very cold with snow and icy roads affecting visibility and mobility, Tesla owners were able to test the latest version of FSD Beta in such harsh conditions as low visibility and slippery roads. Some of these experiences were recorded by Tesla owners the videos of which you can watch below.

To improve the handling of low visibility scenarios such as rain and snowfall, Tesla says that it has upgraded its object detection network. The algorithm of the new Tesla Vision object detection works based on photon counts coming from the video stream.

Remember that Tesla just recently removed ultrasonics after radar from its cars in favor of Tesla Vision. But it seems the automaker is now going to install high-definition (HD) radars into its cars (the cars missing a radar will get a retrofit for sure, most probably without cost but we will have to wait for the confirmation).

With every new version, Tesla is also improving the safety of vulnerable road users (VRUs) such as pedestrians, bicycles, bikes, animals, etc. Tesla first started putting out information on VRU improvements in FSD Beta v10.4 last year.

Video: Tesla FSD Beta performance in the snow using version 10.69.3.3 (one previous version than the current 10.69.25).

FSD Beta v10.69.25 Release Notes (2022.44.25.5)

Upgraded the Object Detection network to photon count video streams and retained all parameters with the latest auto-labeled datasets (with a special emphasis on low visibility scenarios).
Converted the VRU Velocity network to a two-stage network, which reduced latency and improved crossing pedestrian velocity error by 6%.
Converted the Non-VRU Attributes network to a two-stage network, which reduced latency, reduced incorrect lane assignment of crossing vehicles by 45%, and reduced incorrect parked predictions by 15%.
Reformatted the autoregressive Vector Lanes grammar to improve precision of lanes by 9.2%, recall of lanes by 18.7%, and recall of forks by 51.1%. Includes a full network update where all components were re-trained with 3.8x the amount of data.
Added a new “road markings” module to the Vector Lanes neural network which improves lane topology error at intersections by 38.9%.
Added a new “road markings” module to the Vector Lanes neural network which improves lane topology error at intersections by 38.9%.
Upgraded the Occupancy Network to align with road surface instead of ego for improved detection stability and improved recall at hill crest.\
Reduced runtime of candidate trajectory generation by approximately 80% and improved smoothness by distilling an expensive trajectory optimization procedure into a lightweight planner neural network.
Improved decision-making for short deadline lane changes around gores and richer modeling of the trade-off between going off-route vs trajectory required to drive through the gore region.
Reduced false slowdowns for pedestrians near crosswalks by using a better model for the kinematics of the pedestrian.
Added control for more precise object geometry as detected by general occupancy network.
Improved control for vehicles cutting out of our desired path by better modeling or their turning / lateral maneuvers thus avoiding unnatural slowdowns.
Improved longitudinal control while offsetting around static obstacles by searching over feasible vehicle motion profiles.
Improved longitudinal control smoothness for in-lane vehicles during high relative velocity scenarios by also considering relative acceleration in the trajectory optimization.
Reduced best-case object photon-to-control system latency by 26% through adaptive planner scheduling, restructuring of trajectory selection, and parallelizing preception compute. This allows us to make quicker decisions and improves reaction time.
Introduced foundational support for model-parallel neural network inference by sharing intermediate tensors across SOCs to improve road edge and road line prediction consistency through changes to the TRIP compiler, inference runtime, and inter-processor.
Improved handling of traffic control behavior in dense intersections areas by improving the association logic between traffic lights and intersections.

FSD Beta 10.69.2 Testing Videos

Chuck Cook is one of the first ones to review most of the FSD Beta releases. He also got his hands on the FSD Beta 10.69.25 update early and took it for a drive to his routine test route, the Memorial Park drive in Jacksonville, Florida.

Tesla declared in the last point of the FSD Beta v10.69.25 release notes above that this update has improved the intersection + traffic lights behavior in this update. However, as we watch Chuck Cook’s testing video below, we can see that the double green light false slowdown still persists.

Chuck first reported the double green traffic light false slowdown issue in the FSD Beta 10.69.3.1 update but it is still not totally fixed in v10.69.25. As the double green light appeared, his Tesla Model 3 running on FSD Beta slowed down the car from 45 mph to 38 mph.