The software update has been rolled out in North America and mainly includes fixes for recall issues, as well as an update to FSD Beta version 11.3.2. Compared to previous releases, this update has a wider range of users.
- FSD Beta is now available on highways, providing a unified vision and planning stack for both highways and non-highways, and replacing the traditional highway stack that has been in use for over four years. Tesla’s traditional highway stack relies on a single camera and frame network, and can only handle simple specific lane maneuver logic, whereas FSD Beta can use a multi-camera network and next-generation planner to achieve more complex agent interactions, reduce lane dependency, and enable more intelligent behavior, smoother control, and better decision-making.
- By obtaining more data from the autonomous driving test fleet, the recall rate for close-range cut-in situations has increased by 15%, especially for large trucks and high yaw rate scenes. In addition, FSD has also extended and adjusted the dedicated speed control for cut-in objects.
- Improved positioning on wide lanes to allow other cars to pass by leaning towards the upcoming turning direction.
- Improved handling in high curvature or large truck scenarios by offsetting in the lane to maintain a safe distance from other vehicles on the road and to improve comfort.
- Improved path-blocking behavior when changing lanes in heavy traffic. It now maintains a greater distance on blocked lanes to reduce the probability of being surrounded in crowded traffic.
- Improved lane changes in heavy traffic scenarios by allowing for higher acceleration during the aligning stage. This allows it to more naturally find gaps to overtake cars that are very close in adjacent lanes.
- With the lane guidance module integrated with the latest version of the road edge and line network, it can make turns smoother by improving detection consistency between lane, line, and road edge predictions.
- Improved accuracy in detecting other vehicles’ movement intentions. The accuracy in other vehicle turning cases has increased by 23%, and the probability of mis-detecting a parked vehicle ahead with Autopilot has decreased by 12%.
- Expanded the maximum trajectory optimization range for smoother control on high curvature roads and distant vehicles while driving at high speeds.
- Improved driving behavior when approaching a row of parked cars on narrow lanes, tending to deviate but still staying in the lane instead of changing lanes or slowing down for no reason.
- Improved back-to-back lane changes by better integrating vision-based localization and rough map lane data.
- Added text descriptions to the user interface to inform users of FSD’s planned actions, and improved the visualization of deceleration before the vehicle slows down on the path, displaying the acceleration magnitude during deceleration with a V-shaped rendering at different levels of opacity and speed, and drawing a solid line at the point where the car is about to come to a stop.
- Improved recall rate and accuracy of target detection, with a significant 10% reduction in half-car position error, and 3% and 7% increases in recall rate and accuracy for passing vehicles a hundred meters away, respectively, and a 5% increase in recall rate for motorcycles.
- Reduced false offset around objects on wide lanes and near intersections in low-speed scenarios by improving object kinematic modeling.
- When FSD Beta is activated, the position of the automatic blind spot camera is adjusted to prioritize Autopilot visualization, and you can also drag the camera to save a custom position.
According to recent recalls (US #23V085 and Canada #2023-063), Tesla has improved the following specific behaviors in the FSD Beta version:
- Improved decision-making logic for passing through intersections or stopping before a yellow light, considering the required deceleration time for stopping, the time to enter and cross the intersection, and the distance to cross the intersection before the red light transition. This makes handling yellow lights more natural and humane.
- Improved longitudinal deceleration control mechanism when stopping at a stop sign intersection, making the gear shifting process feel more intuitive and natural.
- Improved Tesla’s speed adjustment logic when entering certain speed zones by allowing early control of detected speed limit signs. The response intensity when decelerating based on detected speed limit signs depends on the current speed and the difference between it and the displayed speed on the detected speed limit sign. Tesla will add a visual effect behind the speed limit icon on the user interface, which will alert the driver when the vehicle’s set speed exceeds 50% of the detected speed limit. Finally, Tesla has removed the option for absolute speed limit offset in FSD Beta, and only percentage-based offset is available.
- Improved behavior in which Tesla may continue straight in a turning lane in certain situations.
