Skip to content
Saved

Tesla pulled Optimus training off the assembly line after workers balked at teaching their replacements

Tesla has scaled Optimus output tenfold this year, and some workers who recorded their own movements to train it say they knew exactly why it existed.

By Robovations2 min read

House art for Robovations dispatches: a robot vacuum on the floor of a newspaper press hall, a web of blank paper running over the press rollers into the folder, folded papers on a conveyor and tied bundles on a pallet.

Tesla has spent 2026 ramping up production of Optimus, its humanoid robot for warehouse and factory work. Output climbed from a few dozen units a week in the second quarter to several hundred a week by August, according to a report from The Information. Tesla’s internal target is more than 1,000 robots a week by the end of the year, with a longer-term goal of 20,000 a week.

Getting there depends on teaching the robot to do things a person already knows how to do. Tesla had assembly-line workers in Texas and California wear motion-capture suits and camera rigs that recorded their movements, then fed the footage to Optimus as training data.

Some of those workers objected. The Information reported that employees complained because they knew the robots were designed to eventually replace them. Electrek and Superpower Daily, both citing that report, described the same reaction: staff performing the recorded motions were, in effect, being asked to document their own job’s disappearance.

Tesla’s response was to move the recording work off the line. The company reassigned it to dedicated data-collection teams. It also opened training hubs in Colorado, Arizona and Florida. The people generating footage for Optimus are no longer the same people whose jobs the robot is meant to eventually take.

That reassignment settles who records the footage. It does not settle what the footage is for. The same report says Optimus still can’t reliably generalize beyond the tasks it has already been shown, and its behavior turns unpredictable outside its training. Learning one new basic task can take several days.

Tesla has logged more than 500,000 hours of training video and wants to double that by year’s end. That scale of data collection assumes the problem is coverage. Not the robot’s capacity to generalize from what it has already seen.

The hardware has its own limits. Each hand and forearm on the current V3 robot carries more than 100 screws and small parts that still have to be assembled by hand. Some hand sensors have proven unreliable enough that Tesla is developing a replaceable sensor glove for next year.

Tesla has missed targets on Optimus before. Musk said in January 2025 that the company would build about 10,000 units in 2025, with thousands doing useful work by year’s end. A year later he acknowledged that zero were.

Whether moving training off the assembly line changes anything about the robot itself, rather than just who records for it, is a question neither report answers. Electrek reported the production figures and hardware details on September 25. Superpower Daily, also citing The Information, reported the worker reassignment the same day.

Published September 27, 2026Send a correction