Tesla Robotax started in geofenced area of Austin Texas. June 22, 2025. It currently is by invitation only according to its web page: https://www.tesla.com/support/robotaxi/getting-started. There’a also a landing page for Robotaxi: https://www.tesla.com/robotaxi
It operates like Uber through a Tesla Robotaxi app. Select a destination on app and taxi arrives and takes you to your destination. The robotaxi service includes a safety monitor seated on the front passenger seat with access to a button to stop the car and to pull over. The passenger via the rear touch screen can request Support and talk to a person or pull over. These safety features are a good while to ensure safe operation until the Unsupervised Self Driving rock solid. The car is a Tesla Model Y with Robotaxi stickers on the sides and back. It operates on camera and without Lidar.
Tesla Robotaxi competes with Waymo which has also started service in Austin. One improtant difference is Waymo is making this available to all Uber users and there is no human safety monitor.
How did it perform?
For majority of rides, Robotaxi worked fined and many commented that it drives like a human driver. There was good reason to include a human safety monitor. YouTube video includes some rides that had issues:
One ride had phantom braking where there was hard breaking for no reason. This can lead to accident if car behind unable to break on time.
In a second serious incident, the human safety monitor had to stop the taxi. The taxi was going towards a delivery truck that was backing up.
Tesla Robotaxi still needs time to get to unsupervised self driving. It needs more driving data real or simulated to cover more potential scenario. With more training, it will get better. The two incident sighted above would probably not have benefited with Lidar. It was not an issue with visibility. Weather was good and the Tesla camera and AI system should have definitely detected the large delivery truck. The issue has to do with how to drive in that situation and the Tesla system made an incorrect decision continuing forward. Now that this incident is logged and entered in the training set, in future incidents, it will know how to behave correctly to minimize collision and maximize safety.
Where Lidar might help is the second scenario is the Lidar in Waymo can operate as a backup system to alert the car that forward collision with SOME object is imminent and halt its forward motion. It might not know it’s a UPS truck but LIDAR sees objects and knows how far it it from the Waymo and would see that the distance separating car from object is shrinking and a relatively simple driving rule could be coded up to avoid collisions.
Tesla Robotaxi has taken a first step. It needs to keep improving. This can be tracked by metrics like interventions per miles and interventions per trips. The Tesla FSD system like other AI systems is trained on data and is not explicitly told how to drive. It needs more training to master how to drive in complex situations. As we have seen with Grok and ChatGPT, AI systems on some tasks can exceed human capability. AI can learn driving too. It’s an open questions when it gets there. Waymo with LIDAR has reached that level of competence. Tesla is trying to gather without LIDAR.