Tesla started deploying the latest installment of its Full Self-Driving Beta software version 10.9 last weekend. FSD Beta 10.9 (firmware version 2021.44.30.10) was first rolled out to Tesla’s internal beta tester employees and a couple of days later to the Early Access Program (EAP) participant Tesla owners.
After Elon Musk posted a list of improvements on Twitter that the automaker had made in its FSD Beta version 9.2 last year, Tesla has made it a norm to jot down a list with each new version’s release notes. Tesla specifically mentions the percentage (%) of improvement that it has made in a specific area of a self-driving feature. However, several of these release notes are too technical for even a Tesla owner but still a good way to report progress to your customers.
FSD Beta 10.9 Release Notes (2021.44.30.10)
- Improved intersection extents and right of way assignment by updating modeling of intersection areas from dense rasters (“bag of points”) to sparse instances. Increased 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 AI accelerator chip, through innovations in the AI compiler stack.
- Upgraded generalized static object network to use 10-bit photon count streams rather than 8-bit ISP tonemapped images by adding 10-bit inference support in the AI compiler stack. Improved overall recall by 3.9% and precision by 1.7%.
- 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 a stop more calm when there is an imminent slowdown nearby.
- Improved gap selection when yielding to a stream of oncoming cars on narrow roads.
Tesla has also made architectural improvements like improving the bit count of neural net images from 8 bit to 10 bit. This is a huge improvement when taking into account the world’s largest self-driving imagery data of billions of miles and counting each day. Architectural upgrades like this will surely help expand the Tesla Autopile FSD horizon in the near and long-term future.
In the previous FSD Beta 10.8 update, we saw that one of the Tesla cars did hang a bit while crossing an intersection. Tesla seems to have focused quite a bit on handling intersections in this update as is imminent by looking at the above release notes.
Tesla Model S Plaid owner and YouTuber Black Tesla tested FSD Beta 10.9 on his car. He drove the car on FSD Beta using the Assertive profile, which is one of the more aggressive modes of Tesla’s self-driving software. There are three driving modes of FSD Beta — Chill, Average, and Assertive — following are the details of these as shown on a Tesla center display.
In this profile, your Model S will have a larger follow-distance and perform fewer speed lane changes.
In this profile, your Tesla car will have a medium follow distance and may perform rolling stops.
In this profile, your Tesla car will have a smaller follow distance, perform more frequent speed lane changes, will not exit passing lanes, and may perform rolling stops.
But looking at the attempt of an unprotected left turn, Tesla still needs to improve a lot in this area. It looked like the car lost its confidence just after it crept forward to take the turn. Let’s watch.
On the other hand, another FSD Beta tester Whole Mars Catalog/YouTube was able to drive his Tesla Model 3 for 70 minutes straight. This drive was performed in the streets of San Francisco. Tesla Neural Net is well-versed with the roads of California because it is the state where most Teslas have been sold and driven historically.
Drove around San Francisco until there was a takeover. Managed to go for about 70 minutes before I had to take over due to feeling like car was getting too close to metal barrier.Whole Mars Catalog/YouTube.