Tesla confirmed on its third-quarter earnings call that it is developing an in-house, AI-based mapping and navigation system, and that the first version to ship it will be FSD v15. Ashok Elluswamy, the company's vice president of AI, was blunt when answering a shareholder question:
"We should not depend on external providers' maps."
The shareholder pointed to three long-running complaints about FSD: inaccurate speed-limit recognition, clumsy parking, and awkward route choices. Elluswamy agreed these are problems and said the in-house maps are aimed squarely at them.
What the new maps will cover
According to what has been disclosed so far, the new system builds its maps from road data collected automatically across the whole fleet. It will annotate travel speeds, stop signs, traffic lights and parking lots, and will also flag potholes, driveways and exits with blocked sight lines, and dangerous intersections. Elluswamy stressed that this is a new architecture, not simply a swap to a different map vendor.
The rollout comes in two steps. FSD v15 will first take over the departure and arrival phases of a trip, and over the following months it will gradually extend to navigation, parking and destination handling. The system will also learn the owner's driving preferences and respond to voice commands, and when parking, drivers will be able to prioritize the nearest open space or an accessible space.
Tesla's current navigation is a patchwork. Google supplies the base map, streets, satellite imagery, points of interest and traffic conditions, while Tesla writes the navigation app and calculates routes itself. It added up to three alternative routes in 2022 and launched "preferred routes" this summer. What is being replaced now is the external data layer at the very bottom.
From "maps are a big mistake" to building its own
Put Tesla's statements about maps side by side and the shift is considerable.
When Autopilot launched in 2015, Tesla advertised its high-definition maps as 100 times more detailed than ordinary navigation maps. By the 2019 Autonomy Day, Elon Musk's line had become that HD lane maps were a "huge mistake" and a "very bad idea", on the grounds that relying on maps makes self-driving brittle. Ever since, Tesla has pitched a vision-only, map-light approach as a selling point.
This time the wording leaves room to maneuver. Tesla did not mention centimeter-level HD maps and emphasized fleet data and an AI architecture instead. The idea resembles Waymo's approach of sending sensor-equipped cars to scan a city first, with one difference: Tesla's production cars carry no lidar, so how detailed the maps can get depends on how much the camera data can reconstruct.
What pushed the change of tone
Outside pressure is also mounting. A Belgian road-safety organization tested a 30 km/h zone in Brussels and found that FSD exceeded the limit on 55% of the road, at an average speed of about 44 km/h. One reason was that the in-car display showed the limit as 50 km/h along most of the route.
The US National Highway Traffic Safety Administration (NHTSA) is also assessing whether to act on reports of Tesla Robotaxis speeding in school zones, and has opened an investigation covering roughly 1,000 Model Y and Cybercab vehicles. It remains at the evaluation stage.
The timetable has slipped as well. FSD without driver supervision has been pushed back to the second quarter of 2027, initially open only to the Robotaxi fleet, with privately owned cars not in the first group.
For owners in China, these maps are of no immediate use. Tesla's FSD push in China has already been a bumpy road, and since the new maps rely on overseas fleet data, they would also have to deal with surveying qualifications and cross-border data rules. Tesla has not said where or when v15 will be pushed out, and that cannot be verified for now.
Sources: DigitalToday, CocoLoop, Qudong, Tesla Q3 earnings call; the Belgian road-safety organization's test report was checked for the share of road where the limit was exceeded and the average speed.