Mecka AI Raises $60M to Train Robots on Human Motion
Mecka AI has raised a $60 million Series B led by Sequoia, with participation from Nvidia and Microsoft's venture fund M12. TechCrunch The company announced the financing on October 7, 2026. Mecka Mecka was founded in 2024. TechCrunch
Mecka collects and analyzes human motion data to train humanoid robots and other kinds of robots. It pays people to record everyday tasks, such as making coffee or fixing cars, while wearing body sensors and using smartphones. TechCrunch The method is designed to stay simple for participants.
On its site, the company states its purpose is to bring Physical AI into real life and says it builds data, evaluation, and deployment infrastructure for robots and embodied AI, or AI tied to a physical body. Mecka That infrastructure, in its description, helps robots learn from real human activity and adapt to real settings, with the stated goal to move robots from lab demos to reliable work.
The company said it is hiring across research, hardware, and operations. Mecka Sequoia previewed the investment on October 6 in an article titled 'Partnering with Mecka: Bringing AI to the Physical World', describing Mecka's mission as helping physical AI labs, robotics companies and end enterprises reach physical AGI. Sequoia TechCrunch had previously reported Mecka was nearing a new round at a $500 million valuation. TechCrunch
The broader context here is where robot performance comes from. Language models scaled on internet text, then human feedback and synthetic data, or computer-generated examples. Grasping and walking have no such corpus. Teleoperation, or remote human control, lab capture, and simulation give useful movement samples but are costly to scale and narrow in setting. Mecka trades precise lab capture for volume and coverage with sensors plus phones, then does the hard cleanup in software.
Looking at what this means for robotics teams, the point to watch is data paired with testing and deployment. Data sellers are common. Repeatable test systems are rarer and often more useful to engineers. A program that lifts a mug in one kitchen and fails in the next lacks more than data. It faces variation and calibration issues that need repeated testing. If Mecka ties repeatable tests to its own data, plus a path to run on robots, it offers a development loop, not a static dataset. That will matter to humanoid developers and to buyers who need reliability figures before use near customers or production lines.
In my view, Sequoia's talk of physical AGI is best read as market signaling, not a technical milestone. The near-term test is concrete. Can human motion, captured without robot hardware in the loop, transfer to different bodies with different mass, strength limits, and touch. That retargeting and safety work remains unsolved at scale. The longer-arc hope, which I share, is that broad human priors shorten that transfer, as web data did for language. Success would be steadily fewer failures on ordinary jobs, the coffee-making and repair work Mecka cites, until supervision drops.
