On November 3, Business Insider released an exposé detailing the rigorous training regimen of Tesla's humanoid robot, Optimus. At Tesla's engineering headquarters, within a glass - walled laboratory, scores of employees are tasked with training the robot by mimicking everyday actions. These actions range from the simple act of lifting a glass to more complex ones like wiping a table and drawing curtains.

During an eight - hour work shift, employees are mandated to repeat each action hundreds of times. To capture every nuance and facilitate Optimus's learning of human behavioral patterns, five cameras are strategically placed on their helmets and backpacks, recording the entire process.

Tesla's CEO, Elon Musk, has high hopes for Optimus, considering it the linchpin of the company's future business. He projects that Optimus will contribute approximately 80% of the company's valuation in the long run, with ambitious plans to eventually churn out one million units annually.

The training scope is extensive, encompassing factory operations, household chores, and caregiving tasks. However, the current training process still heavily relies on manual data collection. Employees are required to execute a variety of actions, including sprinting, squatting, and dancing. Some tasks demand repeated refinements as they are deemed "not human - like enough."

On the technological front, Tesla made a significant shift in June. It adopted a pure vision - based training approach, discarding motion capture suits in a bid to expedite data collection. Additionally, it introduced cutting - edge tools such as multi - camera panoramic shooting and haptic gloves.

Yet, this high - intensity work environment has taken a toll on the employees. Physical injuries, such as back and neck strains, have become a common occurrence. The uneven weight distribution of the equipment has even forced some employees to take sick leave.

Despite Musk's assertion that Optimus can already autonomously guide visitors, the reality during actual testing is quite different. When performing bending tasks, the robot still has a 50% probability of toppling over. Moreover, it requires supports to maintain balance over short distances.