After a long time, Tesla updated its software with the latest update. As per the latest reports, Tesla FSD Beta v11.4.1 has been rolled out for North American users. This update improve cornering control and overall smoothness by improving the geometry, curvature, position, type and topology of lanes, lines, road edges and confined spaces. Through improvements such as expanding and cleaning the training data set and updating the lane guidance module, the perception of lanes in urban streets has improved by 36%, intersection scenes have improved by 44%, merge scenes have improved by 27%, and turn scenes have improved by 16%.
However, the latest update adding lane guidance inputs to the occupancy network improves the detection of long-range road features, reducing the median false positive rate by 16%. You can check the complete details which is mentioned below.
Check the Tesla FSD Beta v11.3.2 (2022.45.11) Release Note:
- Improve cornering control and overall smoothness by improving the geometry, curvature, position, type and topology of lanes, lines, road edges and confined spaces. Through improvements such as expanding and cleaning the training data set and updating the lane guidance module, the perception of lanes in urban streets has improved by 36%, intersection scenes have improved by 44%, merge scenes have improved by 27%, and turn scenes have improved by 16%.
- Adding lane guidance inputs to the occupancy network improves detection of long-range road features, reducing the median false positive rate by 16%.
- Strengthen its own ability to judge the initiative to cross pedestrian roads/zebra crossings, and only pass when they can pass safely and quickly, thereby improving the driving efficiency and safety of the car.
- Increased motorcycle recognition rate by 8%, and increased the accuracy of vehicle detection to reduce false positives. In addition, the new model also increases the robustness to visual frame rate changes.
- Thanks to the new framework, it can reduce by 43% the intervention caused by other vehicles cutting into the driving lane of the self-driving car. The framework can do this by probabilistically predicting objects that may cut into the lane and aggressively taking actions such as offsetting and/or adjusting speed to optimize the self-driving car’s positional relationship.
- Improved cut-in control can reduce lane speed errors of adjacent vehicles by 40-50%.
- By using additional features of lane change trajectories to improve the supervision function, the recognition rate of partial lane encroachment by objects is increased by 20%, the recognition rate of high-speed cut-in is increased by 40%, and the recognition rate of cut-out is increased by 26%.
- 68,000 videos were added to the training data set, and the improved automatic labeling of real data was used to further reduce the misjudgment of low-speed highways and improve the accurate estimation of the speed of distant objects.
- By adjusting the magnitude of lateral acceleration control, the lane deviation of large vehicles is smoothed to achieve a more stable driving effect.
- Improved lateral control, further increasing the vehicle’s ability to control an upcoming high-curvature merge, making it more inclined to stay away from the merge.