Photo: @JilianneParker/Twitter

Tesla will start rolling out FSD Beta V10.9–which is getting closer to a single stack—later this week, says Elon Musk. The release note indicates significant improvements in the architecture of the feature, leading to improvements while active.

According to Elon Musk, FSD Beta is considered not so good, in part because it still uses a separate stack for city streets and highway traffic. Moving to a single FSD stack requires a huge amount of neural network training, which Tesla's team and beta testers are currently doing. According to the head of the company, V11 will have a single software stack that will combine city and highway, as well as many other architectural upgrades. On the way to achieving this goal, FSD Beta must make several more important changes. Later this week, the company will begin rolling out the V10.9, which features some architectural improvements that allow the car to drive more smoothly, especially when making difficult, unprotected left turns and overtaking maneuvers.

10.9 going out later this week

— Elon Musk (@elonmusk) January 17, 2022

According to the update, in terms of architecture, Tesla improved intersection extents and right of way assignment by updating modeling of intersection areas from dense rasters (“bag of points”) to sparse instances. The update increases intersection region IOU by 4.2%. The sparse intersection network is the first model deployed with an auto-regressive architecture that runs natively with low latency on the TRIP Al accelerator chip, through innovations in the Al compiler stack.

In addition, the generalized static object network was upgraded to use 10-bit photon count streams rather than 8-bit ISP tonemapped images by adding 10-bit inference support in the Al compiler stack. Overall recall was improved by 3.9% and precision by 1.7%.

Thus, according to the release note, FSD Beta V10.9 received the following improvements:

• Made unprotected left turns across oncoming lanes more natural by proceeding straight into intersection while yielding, before initiating the turn.
• Improved lane preference and topology estimation by 1.2% with a network update and a new format for navigation clues.
• Improved short deadline lane changes with better modeling of necessary deceleration for maneuvers beyond the lane change.
• Improved future paths for objects not confined to lane geometry by better modeling of their kinematics.
• Made launches from stop more calm when there in an imminent slowdown nearby.
• Improved gap selection when yielding to a stream of oncoming cars on narrow roads.

© 2022, Eva Fox | Tesmanian. All rights reserved.

_____________________________

We appreciate your readership! Please share your thoughts in the comment section below. 

Article edited by @SmokeyShorts, you can follow him on Twitter

Follow @EvaFoxU