Tesla's Austin robotaxi launch is weeks away but hasn't started driverless testing

Summary

Tesla is weeks away from launching a robotaxi service in Austin but has not begun driverless testing. As of last month, the company was still running supervised rides with human safety drivers behind the wheel, according to an engineer close to testing and a former employee.

Musk is betting Tesla's future on robotaxi ambitions. The service will initially use existing Model Y vehicles before shifting to custom Cybercabs—essentially Model 3s without steering wheels—that Tesla plans to manufacture next year. At an investor call this month, Musk said the Austin launch could start with as few as 10 Model Y vehicles. Tesla is hiring teams to work remotely using VR headsets to supervise cars when they encounter problems. This teleoperation model is standard in the industry but signals Tesla is not yet at full autonomy for the launch.

Tesla's approach diverges sharply from Waymo's established playbook. Waymo spends roughly a year per city, using sensor data from test cars to build detailed maps that let vehicles navigate safely. The company tested Austin for six months with safety drivers, then six months without them. Tesla trains cars primarily on data from existing drivers using Full Self-Driving, the semi-autonomous mode launched publicly in 2022 that now runs on more than 2.4 million cars. This gives Tesla a massive data advantage. Tesla's system uses only cameras, no lidar or radar, and relies on a neural network trained through on-policy learning. Whenever a human takes the wheel or applies brakes, that intervention is fed back to the model as negative feedback, teaching the system to drive more autonomously.

This scaling approach puts Tesla on a different path than competitors. Waymo's handcrafted, location-specific maps require new teams and fresh data collection in every city. Tesla's vision-only, data-hungry model theoretically scales by simply deploying cameras and feeding video to its Dojo supercomputer cluster. The strategy has blind spots. Computer vision systems hallucinate obstacles and miss real ones—issues lidar-based systems can catch through redundancy. Autonomy researchers including Missy Cummings at George Mason University argue cameras alone are insufficient. Critics point to common driving scenarios such as sun glare, fog, and airborne dust that federal regulators are now examining. Musk acknowledged last month that Tesla's robotaxi won't work in snow.

Operational challenges are real. Preparing to launch in Austin forced Tesla to confront unglamorous logistics: rerouting when cars get stuck, customer service, emergency response planning. The Texas Department of Public Safety has not yet received an emergency plan from Tesla, though it says it's prepared for the launch. Other robotaxi services have already blocked traffic and failed to pull over. Waymo and competitors have spent years solving these problems. Tesla is months away from encountering them at scale.

The camera-only critique carries less weight than skeptics suggest. Humans drive with only vision. Tesla's massive FSD dataset from millions of vehicles is a genuine edge. Simulation and transfer learning—techniques Tesla has invested heavily in—can generate synthetic training data for corner cases like snow. The Cybercab hasn't been built yet. Tesla could add lidar cheaply if needed. The real question isn't whether cameras can work. It's whether Tesla can operationalize a fleet at the scale and speed Musk is promising, when Waymo's decade-long infrastructure play suggests the timeline is long.