Feather launches $29,990 wheeled humanoid for developers, backed by a $7.6 million round

Feather, a Bay Area robotics startup, launched a wheeled, dual-arm humanoid for developers on September 24, 2026 at a promotional price of $29,990, down from a list price of $32,990. The company has raised a $7.6 million round led by Gradient, with Builder Capital, Geometry, Seed Innovation and Virgo VC participating.

Why it matters: most humanoid makers sell a closed stack of robot plus model. Feather sells the hardware and lets buyers choose the compute and the AI model, which is the bet that developer platforms, not vertically integrated robots, will spread fastest.

How much does the Feather humanoid cost?

The Feather robot costs $29,990 at launch, a promotional price against a $32,990 list price. TechCrunch put the price at about half that of Unitree’s H2 Edu, the developer version of Unitree’s full-size humanoid; the base H2 lists at $29,900 on Unitree’s US shop, almost exactly Feather’s price. Feather is roughly double the entry price of Unitree’s G1, which we covered in our look at Unitree’s price war. The difference is the form: Feather is a bimanual mobile manipulator on a wheeled base, not a legged robot, and it is priced as a work platform rather than a research toy.

Gradient Ventures general partner Darian Shirazi framed the price against labour cost. “You would hire a laborer for $50,000 to $60,000 a year,” he told TechCrunch, adding that a buyer can now purchase a Feather robot instead. That comparison leaves out integration, supervision and the software a customer must build or license, which is where most of the cost of a deployment sits today. For a wider price comparison across makers, see our 2026 humanoid price guide.

What are the Feather robot’s specifications?

The robot has two 7-degree-of-freedom arms, each about 1 metre long, and 23 degrees of freedom in total, according to Forbes. It weighs 105 kg. The torso rises and drops through 60 cm of vertical travel. An omnidirectional wheeled base moves at up to 1.3 m/s, per Humanoids Daily. Two hot-swappable 48-volt batteries give about 10 hours of operation, and a flat battery can be changed without shutting the robot down.

Rated payload is 1.8 kg per arm, with 6.8 kg peak. Buyers choose six-DoF hands or parallel-jaw grippers. Onboard compute is swappable, from NVIDIA Jetson Nano up to Jetson Thor, and the software supports Python and ROS 2, built-in teleoperation including Apple Vision Pro and Meta Quest headsets, and collision detection.

Why is Feather called the “Android of robotics”?

Feather positions itself as an open hardware layer that runs other companies’ AI models, the way Android runs other developers’ apps. Forbes and TechCrunch report that customers can run models from NVIDIA, Skild AI, Physical Intelligence or their own. Co-founder and chief executive Hoa Mai put the pitch bluntly: “You can’t buy a Tesla robot today and develop on top of it.” He told Humanoids Daily that the idea that “just a few companies will automate all this work is actually absurd.”

That strategy only works if foundation models become interchangeable enough to run on third-party hardware. Our comparison of robot foundation models shows several are now offered that way. The risk is the one Android makers faced: the platform owner captures the developer relationship while hardware margins compress.

Who founded Feather and who are its customers?

Feather was founded in 2025 by Hoa Mai and Parsa Bakhtiari. Mai previously co-founded Kind Humanoid, which 1X acquired in January 2025. Bakhtiari, the chief product officer, worked on Tesla Model 3 production. According to TechCrunch, Feather robots have spent the past year in field tests, working as cooks in restaurants and cleaning science labs; Forbes also lists manufacturing among the verticals. The company says revenue has passed $1 million.

The production numbers are modest and specific: 30 units a month from a Bay Area site, with a facility planned for 200 a month. At full promotional price, 200 units a month would be about $72 million a year in hardware sales. That is a ceiling on current plans, not a forecast, but it shows the scale a small team can reach when it avoids legs and custom AI.