The most expensive part of a robot is the part that touches the world. For decades, that part,the dexterous hand,has been a research-grade luxury, costing more than a luxury sedan and requiring a PhD to program. Chestnut Robotics, a Santa Clara startup founded in 2025, is betting that equation is backwards. They are building a robotic hand you can buy, break, and modify, with the explicit goal of getting it into enough hands that the software finally catches up.
Their product, the Aero Hand, is a five-fingered, tendon-driven manipulator. It can lift an iPhone from a table, balance the weight across palm and fingertips, and finger an M5 screw without fumbling [TechEBlog, retrieved 2026]. The more telling detail is what it ships with: open-source CAD files, electronics schematics, and a Python SDK [Humanoid.guide, retrieved 2026]. The structure is fully 3D-printed nylon, and it connects via a straightforward USB 2.0 interface running on an ESP32-S3 controller [Humanoid.guide, retrieved 2026]. This is not a black box. It is, by design, a platform for tinkering.
The wedge is affordability and access
Chestnut’s public positioning is a study in contrasts. They talk about "sub-millimeter accuracy" and "24/7 uptime reliability" for factory work [Chestnut Robotics, retrieved 2026]. But they also describe the Aero Hand as "an accessible platform for labs, startups, and hobbyists" [Humanoid.guide, retrieved 2026]. The wedge is clear: build a capable hand at a cost point that allows for volume, then let a community of researchers and developers write the software that proves its value in industrial settings. It is a bottom-up go-to-market strategy for a notoriously top-down industry.
By open-sourcing the design and control stack, Chestnut is attempting to accelerate the learning flywheel. Every lab that buys a hand to train reinforcement learning models becomes a potential source of innovation and validation [ProductCool, retrieved 2026]. The company retains control of the branded product and, presumably, any future proprietary software or integrated mobile platforms. It is a bet that the hardware commoditization curve for dexterous manipulation is steepening, and the real value will accrue to the company that owns the most widely adopted development platform.
The team behind the tendons
The founders bring a blend of AI and robotics pedigree to the challenge. Joe (Xu) Dong, the CTO, previously headed AI planning technology at XPeng and was a core member of Waymo's foundational large model team [Futunn, retrieved 2026]. Co-founder and CEO Evan Tao rounds out the leadership [RocketReach, retrieved 2026]. This background in autonomous vehicle software,a field that treats perception, planning, and control as a unified software problem,is a relevant signal. It suggests Chestnut views the hand not just as a mechanical assembly, but as a sensor-rich endpoint for a broader "physical AI" stack.
The company is lean, with an estimated 2-10 employees [LinkedIn, retrieved 2026]. They operate with a degree of stealth, having recently rebranded from TetherIA to Chestnut Robotics to reflect a sharper focus on advanced robotics [X, retrieved 2026]. Their $2 million pre-seed round, led by Ponderosa Ventures and Thrive by SVG Ventures, provides runway to refine the hardware and build early developer traction [Global Venturing, retrieved 2026].
| Founder | Title | Key Background |
|---|---|---|
| Joe (Xu) Dong | Co-Founder & CTO | AI planning at XPeng; core member of Waymo's foundational large model team [Futunn, retrieved 2026] |
| Evan Tao | Co-Founder & CEO | Background associated with Chestnut Robotics (previously TetherIA) [RocketReach, retrieved 2026] |
Where the grippers are headed next
A robotic hand in a lab is a research tool. A robotic hand on a mobile platform in a factory is a product. Chestnut appears to be building toward the latter. LinkedIn posts from early 2025 show a wheeled base robot performing SLAM navigation autonomously in an office environment, suggesting development of a full mobile manipulator [LinkedIn, Feb-Mar 2025]. The long-term mission, as stated on their site, is building "a single, learnable robot platform for factory work that is too complex, varied, or expensive to automate" [Chestnut Robotics, retrieved 2026].
The roadmap implied is classic hardware startup: seed the market with a compelling component (the Aero Hand), cultivate a developer ecosystem, then layer on higher-margin software and integrated systems for commercial deployment. The next twelve months will be about moving from promising prototypes to verified use cases. Key milestones to watch include the first commercial integrations of the Aero Hand into other companies' robots, any announcements of paid software modules or services, and the demonstration of their mobile platform performing a specific, repeatable task beyond navigation.
The incumbent to beat
Chestnut is not entering an empty field. The competitive landscape for dexterous manipulation includes well-funded giants and specialized startups.
- Boston Dynamics. The gold standard for dynamic mobility and now, with its Atlas platform, dexterous manipulation. Their approach is vertically integrated, proprietary, and astronomically expensive, aimed at defense and heavy industry.
- Mimic Robotics. A direct competitor focused on dexterous hands for logistics and manufacturing, likely pursuing a more traditional enterprise sales motion.
- Tacta Systems. Another player in the advanced manipulation space, though public details are limited.
Chestnut’s bet is that none of these incumbents can or will pursue the open, affordable, community-driven path. They are competing on a different axis: cost of adoption and speed of iteration. The risk, of course, is that the factories they ultimately want to serve will prefer the turnkey, supported solution from an established player, even at a higher price. Chestnut’s answer will be to point to the volume of real-world testing and software libraries their approach has generated.
A back-of-the-envelope calculation makes the bet tangible. If a research-grade dexterous hand costs $50,000 and a lab can budget for one, progress is slow. If the Aero Hand costs one-tenth of that and a lab can buy ten, the number of parallel experiments,and the rate of software advancement,increases by an order of magnitude. That software, in theory, becomes the moat. For Chestnut to succeed, they don't need to out-engineer Boston Dynamics on day one. They need to get their hardware into enough PhD projects and startup garages that the collective intelligence of those users builds something an incumbent cannot easily replicate.
The company they must beat is not the one with the most impressive lab video. It's the one that owns the procurement process for Fortune 500 automation lines. To get there, Chestnut first has to win the workbench.
Sources
- [Chestnut Robotics, retrieved 2026] Company website | https://chestnut.bot/
- [LinkedIn, retrieved 2026] Company profile | https://www.linkedin.com/company/chestnutbot
- [X, retrieved 2026] Rebranding announcement | https://x.com/TetherIA_ai/status/2032330051707146679
- [Humanoid.guide, retrieved 2026] Aero Hand Open product page | https://humanoid.guide/product/aero-hand-open/
- [TechEBlog, retrieved 2026] Aero Hand Open demonstration article | https://www.techeblog.com/open-source-aero-hand-diy-robotic-hand-tetheria/
- [ProductCool, retrieved 2026] Aero Hand Open use cases | https://www.productcool.com/product/aero-hand-open
- [Futunn, retrieved 2026] Founder background article | https://www.futunn.com/
- [RocketReach, retrieved 2026] Team data | https://rocketreach.co/
- [LinkedIn, Feb-Mar 2025] SLAM navigation demonstration post | https://www.linkedin.com/company/chestnutbot
- [Global Venturing, retrieved 2026] Pre-seed funding report | https://news.crunchbase.com/venture/biggest-funding-rounds-ai-robotics-figure/