Tesla has begun building its own maps to address major problems in its self-driving system, Full Self-Driving (FSD).

On Oct. 22 (all times local), Tesla and foreign media reported that Tesla said on its conference call for third-quarter 2026 earnings that it is developing an in-house maps and navigation system to improve speed-limit recognition, parking and route selection.

The move contrasts with Tesla’s previous stance that self-driving systems become vulnerable if they rely on maps. In a shareholder Q&A, an individual investor said FSD shows problems with speed limits, parking and general route selection, and asked why Tesla does not build its own maps instead of using external maps.

Ashok Elluswamy (애쇼크 엘루스와미), Tesla’s vice president of artificial intelligence (AI), replied, "We should not depend on external providers' maps." He said Tesla is building its own AI maps and navigation architecture based on data collected by the entire Tesla vehicle fleet. The system is designed to learn individual driver preferences and respond to voice commands.

Elluswamy said the technology will first handle the start and end points of driving in FSD v15, and then over the following months will take over navigation, parking and destination handling.

Tesla has pursued in-house map development before. At the launch of Autopilot in 2015, it said it was using data collected from customer vehicles to build high-definition maps that were 100 times more detailed than general navigation maps. But at Autonomy Day in 2019, Elon Musk criticised reliance on high-definition lane maps as a "huge mistake" and a "very bad idea," saying it makes self-driving systems vulnerable. Tesla at the time stressed an approach that it could drive anywhere using only cameras and AI.

The problems Tesla is trying to address with maps are tied to its ability to accurately understand road conditions. In particular, FSD has shown problems recognising speed-limit signs. When a Belgian road safety group last month tested a 30 km per hour limit zone in Brussels, FSD sped in 55% of the area and its average speed was 44 km per hour. One reason cited was that the car’s screen showed the speed limit as 50 km per hour across much of the zone.

The U.S. National Highway Traffic Safety Administration (NHTSA) is also reviewing whether to open an investigation into about 1,000 Model Y and Cybercab vehicles over claims that Tesla robotaxis speed in school zones where they are operating.

The plan means Tesla vehicles will be used not only to collect driving data but also to build maps for autonomous driving. It reflects an approach in which information collected by vehicles is fed into Tesla’s own maps to improve driving performance across the fleet, similar to Waymo’s approach of creating city maps with sensor-equipped vehicles before launching services and updating them during operation. A key difference from Waymo is whether vehicles have lidar installed.

Tesla robotaxis are also operating in limited, city-specific geofenced areas. Tesla has not mentioned building centimetre-level high-definition maps, but it could continue refining its maps with data accumulated by its fleet.

Tesla has pushed back the timing for applying FSD without driver intervention to the second quarter of 2027. It also limited the initial scope to the robotaxi fleet rather than all customer vehicles. As Tesla introduces its own maps after criticising map reliance, the gap between its previous plan to implement general-purpose autonomous driving using only cameras and AI and its actual operating method is coming back into focus.