Tesla's Full Self-Driving Software FSD V13 Begins Rolling Out: Code Rewritten, Clean and Simple Comparable to Raptor 3 Engine

Beijing time, December 2 news, Tesla's Vice President of Artificial Intelligence Software Ashok Elluswamy tweeted on Saturday evening local time that the 13th version of Tesla's highest-level driver assistance software, Full Self-Driving (FSD), has begun rolling out to customers.

The release notes for FSD (Supervised) V13 indicate that this update includes several key improvements compared to previous versions. Although the V13 version currently seems to be optimized for vehicles equipped with AI 4 hardware.

Tesla's autonomous driving engineer Arek Sredzki pointed out that the improvements in this update are significant, noting that the system's end-to-end network now allows vehicles to transport passengers from one parking space to another (P2P).

Tesla AI team member Yun-Ta Tsai mentioned that FSD V13 "may be the biggest rewrite since we embarked on our 'photon counting' journey four years ago." He also emphasized that FSD (Supervised) V13 is the result of collaboration among multiple teams at Tesla, including data, vision, compiler, system, firmware, UI, QA, and PM.

In a later post, he clarified that FSD V13 is a complete rewrite. When asked whether the leap from FSD V12 to V13 is greater than the leap from FSD V11 to V12, this Tesla AI team member pointed out that the build of V13 is very similar to SpaceX's Raptor engine V3 designed for the Starship.

"If we can imagine FSD v13, it looks a lot like Raptor v3. Very clean," this Tesla AI team member wrote in a pinned post on X.

Tesla CEO Elon Musk directly stated "Awesome!" when retweeting Elluswamy's tweet. He previously mentioned that the V13 version of FSD is expected to be over 5 times better than V12.5.5.3.

FSD can currently complete most driving tasks but still requires human supervision. Musk stated in November that FSD will outperform human drivers by early 2025 and plans to launch a robotaxi service in 2025.

Investors have been closely monitoring the developments of FSD, as it relates to the timeline for the launch of Tesla's Robotaxi service.

Although the iteration speed of FSD is very fast, achieving truly unmanned vehicles that do not require remote monitoring may have to wait until at least version v14, to achieve a disengagement ratio lower than that of humans.

Currently, Tesla is making extensive preparations for the launch of the Robotaxi service.

FSD (Supervised) V13.2 Update

FSD (Supervised) v13 has upgraded various parts of the end-to-end driving network.

Including:

• 36 Hz, full-resolution AI4 video input

• Native AI4 input and neural network architecture

• 4.2 times data expansion

• 5 times training computation expansion (achieved by Cortex cluster)

• Photon-to-control latency reduced by 2 times

• Speed curves for city streets and highways

• Start FSD (Supervised) from park mode with just a button press

• Integrated start-stop, reverse, and parking functions

• Improved collision avoidance reward prediction

• Improved camera cleaning

• Redesigned controller for smoother and more accurate tracking

• Dynamic planning of closed road routes, displaying closed roads along affected routes when the fleet detects closed roads

Upcoming improvements:

• Model size expansion by 3 times

• 3 times model context length expansion

• Audio input better handling of emergency vehicles

• Improved navigation reward prediction

• Improvement in parking lot false braking and low-speed driving phenomena

• Support for destination options, including curbside parking, parking spaces, lanes, or garages

• Effective presentation of map and navigation input

Improved handling of camera occlusion

Source: Intelligent Hyperparameters, Original Title: "Tesla's Full Self-Driving Software FSD V13 Begins Rollout: Code Rewritten, Clean and Simple Comparable to Raptor 3 Engine"

Risk Warning and Disclaimer

The market has risks, and investment requires caution. This article does not constitute personal investment advice and does not consider the specific investment goals, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. Investment based on this is at their own risk