Skip to content
Bud's Wiki

Chestnut Robotics launches Aero UMI + Aero Hand — a co-designed capture rig + high-DOF dexterous hand system

TL;DR

Section titled “TL;DR”

Evan Tao (@Evan_Tao_T) announces Aero UMI and Aero Hand at IROS 2026 (Chestnut Robotics, booth 842) as a single co-designed system: a wearable UMI-style capture rig and a high-DOF dexterous hand built by the same team to the same spec, so demonstrations recorded through the rig map 1:1 onto the hand at execution time. The framing critique is explicit — prior UMI variants scaled data collection without a robot in the loop, but “none was co-designed with a dexterous hand at this DOF,” so recorded data never fully mapped to what the hand could do. Landing page: chestnut.bot. Launch surface only — no DOF count, BOM, or empirical results disclosed in the tweet thread.

Key claims

Section titled “Key claims”
  • Aero UMI + Aero Hand are launched as one system — the same team designed both to a shared spec so teleop follows the wearer 1:1 rather than through retargeting guesswork [tweet 3/].
  • Prior UMI variants “proved you can collect manipulation data without a robot in the loop” and many variants followed, but none was co-designed with a dexterous hand at this DOF, so captured data never fully maps to hand capabilities [tweet 1/].
  • The Aero Hand is a high-DOF dexterous hand (specific DOF count not stated in the thread) [tweet 0/].
  • Booth 842 at IROS 2026 (through Wednesday); full materials at chestnut.bot [tweet 2/].
  • Reply from @FrameBleed (an independent observer): “Best ergonomics I’ve seen on dexterous hands” [reply thread].

Method

Section titled “Method”

The tweet thread describes only the framing, not internals. The system fits a hardware-first recipe for the human-to-robot retargeting problem — the same lineage as mimic’s M1+U1 (Solving Dexterity: A Full-Stack Approach (mimic hand M1 + wearable U1)) and X Square’s TwinDEX (TwinDEX: A Scalable Path from Robot-Free Data Collection to Real-World Dexterous Manipulation) — where kinematic invariance between demonstrator and target embodiment is enforced by building both endpoints of the pipeline together rather than by post-hoc retargeting math, a reduced action space, or a scaled dataset. The distinctive positioning claim relative to prior UMI (parallel-gripper) work is that co-design has to extend to a dexterous hand at high DOF to be useful, since a low-DOF collection tool cannot capture the space of trajectories a high-DOF hand can execute. No policy, dataset, backbone, or evaluation is disclosed at launch.

Results

Section titled “Results”

None disclosed. Announcement surface: tweet-level engagement (4,614 views, 8 replies, 55 likes, 25 bookmarks at capture time), plus a positive independent reply on ergonomics. All quantitative claims (DOF, BOM, latency, success rates, workspace accuracy) are absent from the launch thread.

Why it’s interesting

Section titled “Why it’s interesting”

Direct sibling in position to Solving Dexterity: A Full-Stack Approach (mimic hand M1 + wearable U1) (mimic M1 hand + U1 wearable, 15-DoF), TwinDEX: A Scalable Path from Robot-Free Data Collection to Real-World Dexterous Manipulation (TwinDEX, bimanual three-finger nine-DoF), and Getting a Grip on Robotic Data Collection — Co-Designed Koala Gripper for Force-Mediated Manipulation (RAI Koala co-designed handheld + robot-mounted gripper) — all four take the hardware-first co-design branch of the Human-to-Robot Retargeting interface-location debate, but Chestnut is (per its own framing) the first to push the co-design constraint into a dexterous hand explicitly co-scaled with the wearable. Contrasts with the scale-first UMI recipes (Xiaomi-Robotics-1 (XR-1) — Scaling VLA Foundation Models with 100K Hours of Embodiment-Free UMI Pre-training XR-1’s 100k UMI hours, HandUMI — Robot-free bimanual data collection retargetable to any parallel-gripper arm (Brik H. Meza launch tweet) HandUMI’s four-arm swap-in-gripper approach) which stay in parallel-gripper embodiments and fix invariance at the collection tool rather than the matched pair. Note: the product name “Aero Hand” collides with TetherIA’s earlier Aero Hand Open: low-cost, open-source, lightweight, anthropomorphic dexterous hand Aero Hand Open ($314 BOM, 7 motors / 16 joints, tendon-driven), a distinct project from a different team — same name, different lineage.

See also

Section titled “See also”