– Enabled FSD Beta on highway. This unifies the vision and planning stack on and off-highway and replaces the legacy highway stack, which is over four years old. The legacy highway stack still relies on several single-camera and single-frame networks, and was setup to handle simple lane-specific maneuvers. FSD Beta’s multi-camera video networks and next-gen planner, which allows for more complex agent interactions with less reliability on lanes, makes way for adding more intelligent behaviors, smoother control and better decision making.
– Improved recall for close-by cut-in cases by 15%, particularly for large trucks and high-yaw rate scenarios, through an additional 30k auto-labeled clips mined from the fleet. Additionally, expanded and tuned dedicated speed control for cut-in objects.
– Improved the position of ego in wide lanes, by biasing in the direction of the upcoming turn to allow other cars to maneuver around ego.
– Improved handling during scenarios with high curvature or large trucks by offsetting in lane to maintain safe distances to other vehicles on the road and increase comfort.
– Improved behavior for path blocking lane changes in dense traffic. Ego will now maintain more headway in blocked lanes to hedge for possible cans in dense traffic.
– Improved lane changes in dense traffic scenarios by allowing higher acceleration during the alignment phase, This results in more natural gap selection to overtake adjacent lane vehicles very close to ego
– Made turns smoother by improving the detection consistency between lanes, lines and road edge predictions. This was accomplished by integrating the latest version of the lane-guidance module into the road edge and lines network.
– Improved accuracy for detecting other vehicles’ moving semantics. Improved precision by 23% for cases where other vehicles transition to driving and reduced error by 12% for cases where Autopilot incorrectly detects its lead vehicle as parked. These were achieved by increasing video context in the network, adding more data of these scenarios, and increasing the loss penalty for control-relevant vehicles,
– Extended maximum trajectory optimization horizon, resulting in smoother control for high curvature roads and far away vehicles when driving at highway speeds.
– Improved driving behavior next to row of parked cars in narrow lanes, preferring to offset and staying within lane instead of unnecessarily changing lane away or slowing down.
– Improved back-to-back lane change maneuvers through better fusion between vision-based localization and coarse map lane counts.
– Added text blurbs in the user interface to communicate upcoming maneuvers that FSD Beta plans to make. Also improved the visualization of upcoming slowdowns along the vehicle’s path. Chevrons render at varying opacity and speed to indicate the slowdown intensity, and a solid line appears at locations where the car will come to a stop.
– Improved the recall and precision of object detection, notably reducing the position error of semi-trucks by 10%, increasing the recall and precision of crossing vehicles over 100m away by 3% and 7%, respectively, and increasing the recall of motorbikes by 5%. This was accomplished by implementing additional quality checks in our two million video clip auto-labeled dataset.
– Reduced false offsetting around objects in wide lanes and near intersections by improving object kinematics modeling in low speed scenarios.
– Adjusted the position of Automatic Blind Spot Camera when FSD Beta is active to prioritize the Autopilot visualization. Drag the camera to save custom positions.
