Tesla Full Self-Driving Beta version 9.2 was actually due on Friday night this past weekend but it got delayed due to some unexpected issues, “Some unexpected last-minute issues. Should go out in next day or two,” Elon Musk tweeted in response to a beta testers question about its rollout.
Later on, in the same discussion thread, Musk posted the following list of improvements that the Silicon Valley-based automaker has made in FSD Beta V9.2 over the previous version 9.1.
FSD Beta 9.2 technical release notes (2021.12.25.15)
- Clear-to-go boost through turns on minor-to-major roads (plan to expand to all roads in V9.3).
- Improved peek behavior where we are smarter about when to go around the lead vehicle by reasoning about the causes for lead vehicles being slow.
- v1 of the Multi-model predictions for where other vehicles expected to drive. This is only partially consumed for now.
- New Lanes network with 50k more clips (almost double) from the new auto-labeling pipeline.
- New VRU velocity model with 12% improvement to velocity and better VRU clear-to-go performance. This is the first model trained with “Quantization-Aware-Training,” an improved technique to mitigate int8 Quantization.
- Enabled Inter-SoC synchronous compute scheduling between vision and vector space processes. Planner in the loop is happening in v10.
- Shadow mode for new crossing/merging targets network will help you improve VRU (vulnerable road users) control.
The above list of improvements that Elon Musk personally posted on Twitter were not part of the FSD Beta 9.2 release notes downloaded with this software update (refer to V9 in-car release notes for details).
FSD Beta 9.2 test videos
Chuck Cook is one of the FSD Beta testers who happens to be an aeronautical engineer and a US Navy commander. Due to his background in engineering and aeronautics, he is able to comprehend Tesla Autopilot behavior differently compared to other beta testers.
Turn Testing and Acceleration Analysis
Here, Chuck is testing the latest build of the FSD Beta i.e. version 9.2 in his Tesla Model 3 to check if it has gained more confidence and improved decision-making while turning the vehicle.
“This is a follow-up test with many examples of unprotected left turns, unprotected forward-facing left turns, and looking closer into the new acceleration behavior introduced in version 9.2,” Chuck wrote in the video description.
Chuck made Tesla Autopilot go through the same unprotected left turn 3 times and as I saw the video closely, the car got better at completing the turn after the first attempt. However, it was a low-traffic scenario, so it is still needed to be tested in a situation with a substantial amount of traffic on the road.
With assistance from the human driver (for now), Tesla Autopilot in FSD Beta 9.2 is able to take unprotected left turns with more confidence, with a little foot-tap on the accelerator pedal for as a go signal. This is giving the Tesla Neural Net some good data points for self-driving training. In the future, the car should be able to take these turns 100% on its own, without hesitation, and safer than a human being.
In another test drive video by Tesla Owners Silicon Valley, the Tesla Model S with FSD Beta V9.2 handles left turns and mountainous roads much better (watch below).
There’s still room for improvement in these scenarios and Tesla is obviously aware of it and the automaker’s AI team is constantly trying to break the code for Full Self-Driving.