Last weekend, Tesla’s VP of AI Software, Ashok Elluswamy, confirmed on X platform that Tesla has begun rolling out version 13 of Full Self-Driving (FSD) to users.
“Amazing!” CEO Musk later retweeted Elluswamy’s post. He had previously claimed that FSD v13 was expected to be over 5 times better than v12.5.5.3.
According to the official release notes, compared to the previous version, the latest v13 provides numerous critical improvements for vehicles on the AI4 (HW4) platform. Tesla Autopilot engineer Arek Sredzki pointed out that the new version has truly achieved the “parking space to parking space” (P2P) functionality.
FSD v13 not only shows significant capability improvements but has also been “slimmed down.” A Tesla AI team member stated that FSD v13 is “completely rewritten,” representing the collaborative effort of Tesla’s data, vision, compiler, systems, firmware, UI, QA, PM, and other teams.
“FSD v13 is as clean and elegant as the Raptor 3 engine.”
▲ SpaceX’s Raptor v3 engine designed for Starship
Engineer Sredzki demonstrated FSD (Supervised) v13.2’s “point-to-point” driving capabilities through a video. The system can be activated when the vehicle is in a parked state. In the video, the vehicle first reverses to create distance from the car in front before maneuvering around.
Throughout the journey, Sredzki’s vehicle encountered no issues. FSD smoothly handled familiar scenarios like unprotected left turns and intersection navigation.
The most impressive part comes at the end of the video, where the vehicle actively enters a parking lot near its destination.
It then finds a parking space in front of the destination and parks, completing a true point-to-point autonomous drive.
A seasoned FSD beta tester @AIDRIVR noted that while FSD 12.5 was already smoother than most Uber rides he’d taken, FSD V13.2 is even smoother, “better than any Uber I’ve ridden in.”
Another beta tester @Chuck Cook also emphasized V13.2’s smoothness, saying the new FSD behaves more like a human driver, especially when making U-turns – if it can’t complete the turn in one go, it will reverse and try again, performing a “K turn.”
By the way, at a Huawei smart driving communication meeting last month, Huawei revealed that this functionality would soon be available on ADS3.0 as well.
Back to FSD, many users who updated to FSD v13.2 found that the new system can achieve zero interventions even in snowy conditions, accurately entering and exiting parking spaces. The point-to-point autonomous driving capability remains unaffected, and it can even autonomously drive into Supercharger stations at night.
Of course, plugging and unplugging the charging cable still needs to be done manually.
Based on feedback from these engineers and users, current FSD can complete most driving tasks, though human drivers still excel in terms of execution efficiency. However, Musk’s expectations for FSD aren’t about speed and efficiency – he has always emphasized “safety.” Last month, Musk stated that FSD would become “better than human drivers” by early 2025.
During October’s earnings call, Tesla reported that throughout 2024, FSD v12.5 successfully increased the MPI (Miles Per Intervention) by two orders of magnitude.
Third-party website FSD Tracker’s data also reflects FSD’s improved stability. Their data shows that after Tesla’s FSD v12 update, the percentage of completely intervention-free trips increased from 47% to 72%, with average MPI rising from 186 km to 536 km.
With FSD v13, Tesla hopes to improve MPI by 5-6 times compared to v12.5, ultimately achieving a three orders of magnitude (1000x) improvement throughout 2024.
▲ FSD’s cumulative mileage
Additionally, Tesla’s MPI improvements will continue into next year, with expectations to surpass human-level performance in Q2 (or latest Q3) 2025. Of course, human drivers don’t have “interventions” – this refers to the probability of traffic accidents caused by human factors.
As everyone knows, Musk unveiled a sci-fi-looking Robotaxi called Cybercab in October. Recently, it has begun appearing in showrooms across Europe.
Tesla has indeed taken a major step toward Robotaxi services, not with Cybercab, but with actual ride-hailing – Tesla has begun offering FSD-based ride-hailing services to Tesla employees working in the San Francisco Bay Area. Employees can hail rides through a dedicated app to be taken anywhere in the Bay Area.
Musk has also stated that when this service opens to the public in the future, passengers’ personal profiles will automatically sync upon entering the vehicle, enabling personalized in-car settings. Multimedia, navigation, and climate control can also be adjusted through the app.
Tesla plans to launch this service to the public in 2025, with Texas being the first state due to its relatively lenient autonomous driving testing requirements. Meanwhile, Tesla is also working to secure approval in California.
However, it’s worth noting that since current FSD’s MPI isn’t yet “better than human drivers” as Musk claimed, these vehicles still have safety operators in the driver’s seat.