Carnegie Mellon University researchers have unveiled Pixel, a wheeled humanoid robot for mobile manipulation that they say costs less than $15,000 to build and reached 82 percent task progress.
The result arrives while far better-funded companies chase general-purpose robots. Pixel's pitch is that a cheap body, designed in tandem with the method used to teach it, could undercut that race on price.
What the researchers built
The work is documented on arXiv under the identifier 2611.03648 and on the project site, pixel-robot.com. The team describes Pixel as a wheeled humanoid whose hardware and learning approach were co-designed instead of developed separately. A companion page covers a related effort called Pixel-UMI, and a second arXiv paper (2611.03647) appears alongside the main one.
The 82 percent figure measures task progress, and the researchers reported it themselves. It is not an independent benchmark. The materials we reviewed for this story did not include task lists, trial counts or comparisons with other robots, so the number can't yet be weighed against anything.
A crowded commercial field
Money and deployments are moving quickly around Pixel. Skild AI published a blog post titled "Skild crosses $100M ARR," and its financing page lists a Series C. Sunday Robotics' Series B is billed as a $165 million raise at a $1.15 billion valuation, though valuations like that reflect the moment of financing, not what a company is worth today.
Some of the machines are already at work. Agility Robotics announced an expanded partnership with the logistics operator GXO to speed humanoid deployment in U.S. operations, and Figure has published a post about its F-03 robot at BMW. GXO has separately announced a warehouse automation partnership with Reflex Robotics.
Rival platforms and models
Competition extends to hardware and software alike. Hello Robot launched Stretch 4, a mobile manipulator, and published a datasheet in May 2026. Physical Intelligence has opened an API beta. Google DeepMind announced Gemini Robotics 2, which it frames around whole-body intelligence, and Nvidia has announced new reference humanoid frameworks, simulation pipelines and data for its Isaac platform.
