Shulou(Shulou.com)11/24 Report--

CTOnews.com February 20 news, Tesla FSD Beta v11.3 complete update log has been released, the focus of the update is to support the implementation of FSD Beta on the highway, unifying the high-speed and non-high-speed visual and regulatory technology stack.

The complete update log attached to CTOnews.com is as follows (blogger @ is not translated by Zheng Xiaokang):

-enable FSD Beta on the highway. This unifies the high-speed and non-high-speed vision and regulation technology stack and replaces the traditional high-speed stack which has a history of more than four years. Traditional high-speed stacks still rely on several single cameras and single-frame networks and are set to handle simple lane-specific operations.

-FSD Beta's multi-camera video network and next-generation planner allow more complex agent interactions while reducing lane dependence, laying the foundation for more intelligent behavior, smoother control, and better decision-making.

-added voice-based feedback. After taking over, you can now send anonymous voice messages to Tesla describing your experience to help improve Autopilot.

-expand automatic emergency braking (AEB) coverage to handle vehicles crossing the vehicle's path. This includes situations where other vehicles run red lights or cross their own roads and seize the right of way.

-before rerunning, such collisions indicate that 49% of the scenarios will be improved by AEB updates. This improvement is now enabled by both manual driving and Autopilot operation.

-by increasing reliance on instantaneous kinematics and trajectory estimation of objects, the response time of Autopilot to red light runners and stop signs is reduced by 500ms.

-A network of long-distance high-speed lanes has been added to be able to respond to traffic jams and high curvature earlier.

-the target pose prediction error of the candidate trajectory neural network is reduced by 40%, and the running time is reduced to 1 prime 3. This is achieved by using higher performance, more robust offline optimization to improve the dataset, quadrupling the size of the improved dataset, and implementing a better architecture and feature space.

-improve occupancy network detection by oversampling 180000 challenging videos, including Rain Water reflections, road debris and high curvature.

-by adding 40,000 auto-tagged fleet clips for this scene to the dataset, the recall rate of close-up plug cases has been increased by 20%. In addition, the motion model of the plug case is improved to improve the handling of the plug case, and the same model is used to control the plug object more smoothly horizontally and vertically.

-added "lane guidance module and perceived loss" to the road edge and lane network, increasing the lane absolute recall rate by 6% and the road edge absolute recall rate by 7%.

-the overall geometry and stability of lane prediction are improved by updating the Lane Guide module representation with information related to predicted intersections and oncoming lanes.

-improve processing in the case of high speed and high curvature by inward lane line offset.

-improved lane changes, including: earlier detection and processing of simultaneous lane changes, better selection of gaps when approaching congested lanes, better integration of speed-based and navigation-based lane change decisions, and more differences between FSD driving profiles and speed lane changes.

-the smoothness of the longitudinal control response when following the front car is improved by better simulating the possible influence of the brake lights of the front car on its future speed curve.

-increased the detection of rare objects by 18% and reduced the depth error of large trucks by 9%, mainly due to the migration to more intensive automatic tagging datasets.

-the semantic detection of school buses is improved by 12%, and the semantic detection of vehicles from static to mobile is improved by 15%. This is achieved by improving the accuracy of dataset annotation and increasing the dataset size by 5%.

-improve the decision-making at the crosswalk by using the self-trajectory estimation based on neural network instead of the approximate kinematic model.

-improves the reliability and fluency of merge control by abandoning traditional merge area tasks to support merge topologies derived from vector lanes.

-unlock longer fleet telemetry clips (up to 26%) by balancing compressed IPC buffers and optimized write scheduling across dual SOC.

CTOnews.com learned that Tesla FSD Beta v11 was originally planned to be launched in November 2022, but it has been delayed many times and has not yet been launched. Musk admitted that the V11 was "much more difficult than expected" because the update brought a "neural network rearchitecture" to the FSD beta, and no car owner has received the update so far.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.