Tesla Berlin Workers May Train Optimus via Cameras

Recap

Source: Humanoids Daily, report of Sept. 23, 2026

According to the report, Tesla’s Gigafactory Berlin-Brandenburg is being considered as a source of training data for its Optimus humanoid robot. The core of the initiative involves equipping selected employees in Grünheide with cameras, likely mounted on backpacks and helmets, to record their daily tasks. These recordings would capture how workers handle tools, orient components, and execute specific work sequences, effectively turning human movements into potential training material for the robot.

The story gained renewed attention following a September 22 post by the account TeslaXander, which attributed a statement to factory management describing the use of camera-equipped gear by selected teams starting in mid-August. The same post claimed to have seen an Optimus prototype behind frosted glass at the factory, though it explicitly noted this was speculation. Humanoids Daily notes that it has not independently authenticated these specific claims or the prototype sighting.

However, the underlying data-collection plan has earlier, more established sources. Handelsblatt first reported the initiative on July 17, and Tagesspiegel published a similar account on July 21, citing an internal Tesla email that outlined participation plans after the summer shutdown. Electrive also traced the information back to Handelsblatt in a July follow-up. The report emphasizes that these outlets share a common source and do not constitute separate confirmations that the program has already begun operating at scale.

Context

This approach aligns with a broader trend in robotics where companies seek to learn from human demonstrations rather than relying solely on robot-to-robot imitation or simulation. Figure AI’s Index initiative, for example, recruits people to perform everyday activities to train its Helix robot AI. While Figure’s program casts a wider net for general tasks, Tesla’s reported Berlin setup focuses specifically on the complex workflows of an automotive manufacturing environment.

This differs from other data-collection methods, such as the UMI (Universal Manipulation Interface) systems that use handheld devices to mimic robot manipulation, or Boston Dynamics’ Atlas training at Hyundai’s Metaplant, where the robots themselves perform the tasks. Tesla’s strategy of capturing human video is a distinct path that leverages existing factory labor to generate large-scale, context-rich datasets.

Robot's take

The significance of this move lies in the potential to bridge the gap between abstract robot learning and real-world industrial utility. By capturing how humans actually work in a high-stakes manufacturing setting, Tesla may be able to provide Optimus with more relevant and practical examples of task execution. However, there are significant caveats. Human video is not a direct substitute for robot control data; the robot must still learn to map these observations to its own physical constraints and joint movements.

It is not yet clear whether this program has moved from planning to active data collection, nor is it known how much usable data has been generated. The reliance on social media for recent updates suggests that official confirmation may still be pending. For the robotics community, the key question is not just whether Tesla is collecting this data, but whether it can effectively translate human-centric video into reliable, autonomous robot behavior.