Tesla Builds Its Own Maps to Fix Driving Glitches

Tesla is developing its own proprietary mapping system to address recurring failures in its Full Self-Driving software, a move that contradicts the company’s long-standing insistence that detailed maps are unnecessary for autonomous driving. The initiative, revealed by the company’s head of AI during the Q3 2026 earnings call, aims to improve how the vehicle handles speed limits, parking maneuvers, and route selection. This development signals a practical shift in strategy, as Tesla’s fleet now performs a function similar to that of Waymo’s vehicles: collecting data to create a digital representation of the road network to compensate for what the vision system cannot reliably determine on its own.

The decision stems from direct feedback from shareholders, who identified mapping as a persistent weakness in the current system. During a question-and-answer session, a retail investor noted that issues with speed limits, parking, and typical routes remained the most common complaints, prompting the query of why Tesla continued to rely on third-party navigation data. Ashok Elluswamy, Tesla’s VP of AI, acknowledged the critique, stating that the company should not depend on external maps. He confirmed that Tesla is constructing a new AI-based mapping and navigation architecture that leverages collective learning from the entire fleet, while also incorporating individual driver preferences and voice command responses.

The rollout of this technology is planned in stages. Elluswamy indicated that the system will initially handle the start and end of trips with the release of FSD version 15. Over the coming months, the new map layer is expected to take over broader navigation, parking, and destination management functions. This phased approach suggests that the new infrastructure is being tested for reliability before being fully integrated into the core driving experience.

This pivot represents a notable reversal for the company, which has historically dismissed the use of high-precision maps. In 2015, when launching Autopilot, Tesla told Fortune it was building high-precision maps with granularity one hundred times greater than standard navigation, using data from customer vehicles. However, by the 2019 Autonomy Day, CEO Elon Musk had shifted the narrative, declaring that relying on high-precision lane maps was a “huge mistake” that made self-driving systems “extremely brittle.” In that same presentation, he famously labeled lidar a “fool’s errand,” asserting that Tesla would solve driving through cameras and AI alone, without the need for detailed cartographic crutches.

Seven years after that declaration, Tesla is now using its fleet to build a map layer specifically to correct errors in its vision system. While the company maintains that this is not a high-definition map in the sense Musk previously criticized, the problems it addresses—such as misreading speed limits and failing to identify parking options—are fundamentally about understanding the physical environment. The need for a dedicated map layer implies that the end-to-end vision model is still struggling with basic environmental cues that a structured map can provide.

Recent evidence supports the need for such improvements. Last month, a Belgian road safety group reported that FSD vehicles exceeded the 30 km/h speed limit in 55% of the zones they tested in Brussels, averaging 44 km/h. A contributing factor was the vehicle’s display frequently showing a 50 km/h limit in those areas. In the United States, the National Highway Traffic Safety Administration is considering an investigation into claims that Tesla Robotaxis speed through active school zones, affecting approximately 1,000 Model Y and Cybercab vehicles. These incidents highlight that if a vision system cannot reliably read and apply a simple speed limit sign, the solution may require more than just better neural networks.

The new approach brings Tesla’s methodology closer to that of its competitors, particularly Waymo. Waymo creates high-definition maps by driving sensor-equipped vehicles through cities before launching service, continuously updating those maps as its fleet operates. Tesla is now describing a similar process, where cars collect visual data that feeds into a Tesla-built map to improve the driving performance of every vehicle in the fleet. The primary hardware difference remains that Waymo’s vehicles are equipped with lidar, while Tesla relies on cameras.

This strategy also aligns with the operational reality of Tesla’s Robotaxi service, which currently functions in specific geofenced areas within cities. Although Tesla is not currently discussing centimeter-level high-definition maps, a fleet-built map that grows in detail over time could easily evolve in that direction. As the company builds this infrastructure, it moves further from the vision of a universal, map-less self-driving solution and closer to the established model of geofenced, map-assisted autonomous driving.

The shift underscores a decade-long journey in Tesla’s self-driving efforts, where the company has repeatedly refined its approach based on real-world performance. While every major self-driving program utilizes maps, Tesla’s previous messaging suggested otherwise. Now, with the company building its own cartographic tools to fix fundamental navigation errors, the distinction between Tesla’s and its competitors’ approaches appears increasingly blurred, leaving only the presence or absence of lidar as a clear technical differentiator.