Tesla's official account announced on social media platform X that a new version of its Full Self-Driving (Supervised) software has begun rolling out. This update increases system response speed by 20%, further enhancing safety, along with numerous other improvements.
The complete release notes are as follows: FSD (Supervised) v14.3 includes: - An upgrade to the reinforcement learning (RL) phase of FSD neural network training, leading to improvements across various driving scenarios. - A complete rewrite of the AI compiler and runtime using MLIR from the ground up, resulting in a 20% improvement in reaction speed and faster model iteration. - A reduction in unnecessary lane drifting and minor tailgating behavior. - Improved decisiveness in parking space selection and maneuvering. - Enhanced prediction functionality for parking spot locations, now displayed on the map with a (P) icon. - Improved response capabilities for emergency vehicles, school buses, law enforcement vehicles, and other rare vehicle types. - Enhanced handling of small animals by focusing reinforcement learning training on more challenging examples and increasing rewards to promote proactive safety awareness. - Better traffic light handling at complex intersections, including those with compound signals, curves, and yellow light stops, benefiting from training on difficult reinforcement learning examples from the Tesla fleet. - Improved management of rare or special objects extending, hanging, or leaning into the vehicle's path by incorporating rare events from the fleet. - Enhanced handling of temporary system faults by maintaining control and automatically recovering without requiring driver intervention, thereby reducing unnecessary disengagements.
Upcoming improvements: - Extending reasoning capabilities to all behaviors beyond just destination handling. - Adding pothole avoidance functionality. - Increasing driver monitoring system sensitivity through improved eye tracking, better compatibility with glasses, and higher accuracy under various lighting conditions.
Comments