Linkerbot teaches LinkerHand O30 dexterous hand to perform card tricks

TL;DR

Linkerbot — a Chinese dexterous-hand vendor whose LinkerHand series is in mass production — posted a 48-second demo of its LinkerHand O30 performing sleight-of-hand card tricks. The tweet is a product-marketing clip rather than a technical disclosure: it shows end-effector capability but does not describe actuation, control stack, or whether the trick execution is teleoperated, scripted, or a learned policy. The clip lands at a moment when dexterous-hand hardware launches are coming on a near-weekly cadence (Unitree Dex5-S, ORCA, Chestnut Aero Hand, Lumos MOS 2, RewardAI Omnibody, Sharpa) and manipulation-of-thin-flat-objects — long a canonical hard case for multi-finger hands — is a visible benchmark the vendors are starting to compete on.

Key claims

The LinkerHand O30 is shown picking up, fanning, and manipulating playing cards on a felt surface in a tabletop demonstration [tweet video].
The tweet frames the demo as a September 2026 internal project (“This September we taught our hands card tricks”) with a “we’ll show you how it was done” follow-up implied [tweet text].
Linkerbot’s bio states the LinkerHand series is now in mass production, positioning this as a commercial-platform demo rather than a research prototype [profile bio].

Method

Not disclosed in the tweet. The video shows the hand operating on a felt table, which rules out the pure-friction baseline (a reply from @andrewmccalip explicitly asks “Can you pull them off the felt?” — a sanity check about whether the manipulation extends to the harder lift-from-flat-surface case). No information is given about the teleoperation / autonomy split, sensing modality (tactile vs. vision-only), or policy training.

Results

Qualitative only. The reply thread registers roughly 6K views in the first 24 hours and generally positive engagement, but there is no quantitative success rate, no comparison to other dexterous hands on the same task, and no eval protocol.

Why it’s interesting

Card manipulation is a useful informal benchmark for the thin-flat-object regime that has historically broken behavior-cloning policies on multi-finger hands — it stresses fingertip precision, low-force contact modulation, and visual tracking of near-planar objects simultaneously. This fits into the broader stream of dexterous-hand launches the wiki has been tracking: Unitree Dex5-S — 22-DoF 1:1 anthropomorphic dexterous hand with backdrivable joints, from $6.5K (Unitree Dex5-S, 22-DoF anthropomorphic hand at $6.5K), ORCA: An Open-Source, Reliable, Cost-Effective, Anthropomorphic Robotic Hand for Uninterrupted Dexterous Task Learning (ORCA open-source anthropomorphic hand), Chestnut Robotics launches Aero UMI + Aero Hand — a co-designed capture rig + high-DOF dexterous hand system (Chestnut Aero Hand co-designed with capture rig), and Aero Hand Open: low-cost, open-source, lightweight, anthropomorphic dexterous hand (Aero Hand Open). Where those launches have focused on hardware specs (DoF count, cost, backdrivability), this clip is the inverse — a capability demo on a mass-production platform with no hardware disclosure — and underscores that the dexterous-hand ecosystem now has enough platforms available that what you can do with them is becoming the differentiator rather than whether you can build one.