CMU Pixel wheeled humanoid costs under $15,000 to build and learns chores from people wearing copies of its grippers
Pixel is a wheeled humanoid robot and learning system from Carnegie Mellon University, led by researchers Tony Tao and Jason Jingzhou Liu and advised by Deepak Pathak and Ruslan Salakhutdinov. Their paper, posted to arXiv on Nov. 2, says the robot can be assembled for under $15,000 and pairs a human-proportioned upper body with 17 degrees of freedom with a wheeled base and vertical lift.
People can teach it by wearing 3D-printed grippers that match the robot's own; the full wearable kit costs under $1,000, per the paper. Policies trained on these human demonstrations reached 82.0% average task progress across three tasks. A companion paper, Pixel-UMI, shows tasks like fetching a drink from a fridge learned from human data alone, with no robot demonstrations.
The project page says Pixel will be open-sourced soon; no code or design files are posted yet.
More on humanoid robots.
About the footage: Lab teaser video from the project page. The team lists tidying a pillow among its autonomous tasks, but the teaser doesn't label each shot as autonomous or teleoperated.
The source post
Post by clankr on X.
View the post on X
Sources
- Pixel project page (Carnegie Mellon University)
- arXiv: Pixel: A Co-Designed Full-Stack Robot System for Learning General-Purpose Mobile Manipulation
- Pixel-UMI project page
- arXiv: Pixel-UMI: Learning Environment-Aware Whole-Body Manipulation from Sensorized Human Demonstrations
- clankr (@clankrmedia) on X
New to robots? Read our guide: How to tell if a robot video is real: CGI, sped up, or teleoperated?.
Summarized by Servospace from the sources above, drafted with AI help and checked against those sources before publishing. Claims are the companies' or outlets' unless we say otherwise. How we work.