On April 30, 2021, Tesla AI lead Andrej Karpathy gave an interview on the Robot Brains podcast about Tesla's transition to "Software 2.0" and FSD efforts. A subsequent article details how Tesla's FSD solves lane connectivity through a vision-based approach. According to a patent application, US 2026/0170852 A1, titled "Vision-Based Machine Learning Model for Lane Connectivity in Autonomous or Semi-Autonomous Driving," Tesla repurposes an autoregressive transformer architecture, typically used for generative AI and LLMs, to predict lane connectivity. The system processes raw camera pixels through backbone networks and multicam fusion via a cross-attention transformer into a bird's-eye view (BEV) vector space, then uses autoregressive transformers to generate a lane graph. This contrasts with competitors like Waymo and Zoox, which rely on high-definition (HD) localized maps.