Tesla releases FSD Beta Version 11.4.1 update featuring significant architectural improvements

On May 11, Tesla rolled out the FSD Beta V11.4.1 update to North American users, featuring significant architectural improvements. According to Musk, this version could theoretically serve as the foundation for the upcoming FSD V12.0 release—though the actual V12 designation will be reserved exclusively for a fully end-to-end AI-driven iteration. Here’s the release note for FSD Beta V11.4.1:
By enhancing the geometry, curvature, positioning, type, and topology of lanes, roadways, road edges, and constrained spaces, we’ve improved steering control and overall ride smoothness during turns. Through enhancements such as expanding and refining the training dataset, as well as updating the lane-guidance module, we’ve achieved a 36% improvement in lane detection on urban streets, a 44% enhancement in intersection scenarios, a 27% boost in merging scenarios, and a 16% improvement in turning scenarios. Additionally, by integrating lane-guidance inputs into the occupancy network, we’ve optimized the detection of long-distance road features, leading to a 16% reduction in the median false-positive rate. The system has also strengthened its ability to proactively assess pedestrian crossings and zebra crossings, allowing vehicles to proceed only when it’s safe and efficient for them to do so—thereby enhancing both driving efficiency and safety. Furthermore, motorcycle recognition accuracy has been increased by 8%, while vehicle detection precision has been refined to minimize false alarms. Moreover, the new model now demonstrates greater robustness against variations in visual frame rates. Thanks to this advanced framework, interventions caused by other vehicles unexpectedly cutting into the autonomous car’s designated lane have been reduced by 43%. The framework achieves this by probabilistically predicting objects likely to enter the lane and proactively responding—whether by adjusting the vehicle’s lateral position or modifying its speed—to ensure optimal spatial relationships and minimize disruptions.

Improved lane-keeping control capabilities can reduce lane-speed errors between adjacent vehicles by 40–50%. By incorporating additional features from lane-change trajectories into the supervision system, we’ve enhanced object recognition accuracy: specifically, identifying partial lane intrusions has become 20% more precise, high-speed lane changes are now detected with 40% greater reliability, and exit maneuvers are recognized with 26% improved accuracy. To further refine performance, we added 68,000 video clips to the training dataset and implemented an advanced automated labeling process for real-world data, significantly reducing instances of misjudging low-speed scenarios on highways while enhancing the vehicle’s ability to accurately estimate speeds of objects moving at greater distances. Additionally, by fine-tuning the magnitude of lateral acceleration control, we’ve achieved smoother lane shifts for larger vehicles, resulting in a more stable driving experience. This improvement also strengthens the vehicle’s ability to handle upcoming high-curvature merge lanes, helping it naturally steer away from such areas rather than being drawn toward them.
Musk previously mentioned that FSD Beta v11.4 performed exceptionally well during testing, achieving "zero disengagements" (to use his exact words, "zero safety-critical interventions") after several days of driving from random addresses in Austin. Currently, Tesla's FSD is already available in North America, with AP users paying a monthly subscription fee of $199 (note: this currently translates to about 1,375 RMB), while upgrading to EAP costs $99. Tesla continues to roll out new software updates for the FSD beta program and steadily expands the pool of test-driving owners. Since the broader public launch of the beta version last year, over 400,000 Tesla owners in North America now have access to FSD Beta testing privileges—though it remains unclear when FSD Beta v11.4 will be rolled out to the general public on a larger scale.

Translated from Sina Auto

Email:
info@adtfm.com

Phone:
021-20532053

Address:
2nd Floor, No. 150 Jindian Road, Pudong New Area, Shanghai