InHandPlus Vibe-Watch · results

Teach a robot from your wrist.

Human-in-the-loop on a smartwatch: show a robot once, confirm with a tap, and it remembers. Here is what that did to a MicroDuck fetching a thrown tissue — measured, not staged.

in simulationthe watch is real2026-09-05
92%
recognizes what you showed it
head-camera recognition after three taps of correction from the wrist
0 of 10
went for the wrong object
when the wrist could answer — the same runs without it: 4 of 10
90%
placed it where you showed
30 runs, 5 landing spots, 6 seeds — thrown, found, picked, placed

InHandPlus What the watch does

Show it. It finds it. It asks when unsure.

Show itin simulation
1

Show it

Pick the thing up wearing Vibe-Watch. Three frames from the wrist camera become a 7 KB memory — no training, no GPU, no app to build.

The robot finds itin simulation
2

The robot finds it

The robot's own camera matches what it sees against your memory and goes: finds it among socks, caps and coins, picks it up, carries it to the spot you showed.

Unsure? It asks your wristin simulation
3

Unsure? It asks your wrist

When two things look alike, the watch buzzes: “this one?” One tap answers. The memory updates, and it never confuses the two again.

InHandPlus Human in the loop

It asked. You answered. Done.

A sock and a tissue look alike to a small camera. Left: the robot's first guess. Right: after one tap from the wrist — on a layout it had never seen. That is the whole idea of Vibe-Watch: the person stays in the loop, one tap at a time.

Robot's first guess: the sock is boxed as tissue
Robot's first guessin simulation
wrist: “no, that's a sock”
After one tap from the wrist: tissue, sock, cap and coin all recognized
After one tap — all four rightin simulation

InHandPlus Throw it. Say it. Done.

From your hand to the spot you showed

You throw a tissue. The duck searches, finds it among other things on the floor, walks over, picks it up, and places it on the mat you showed it once — while the watch confirms the target on the way.

97%
found the thrown tissue
97%
walked over and picked it up
90%
placed it on the spot you showed

30 runs, 5 landing spots · median 46s from throw to placed · memory 7 KB from 3 wrist frames

20-second cut · the duck asks, the wrist answersin simulation

InHandPlus Why a watch

Four things only a wrist can do

1

Hands-free teaching

You teach by doing — picking things up, putting them where they belong. A phone can’t: if you’re holding the phone, you’re not holding the object.

2

No training, no GPU

Nothing is retrained. The watch writes a small memory the robot reads. Adding a new object is three frames; a new spot is one record.

3

Human in the loop

The robot asks when it isn’t sure, and your answer becomes part of the memory. It gets better with every tap instead of every firmware update.

4

Works on any robot’s built-in skills

Vibe-Watch doesn’t replace how a robot walks or grips. It tells the robot what and where — the layer no robot ships with today.

InHandPlus Get it first

After the nine presets, there is something to do.

Vibe-Watch ships as a bundle companion for MicroDuck. Waitlist members get first access to test point-to-fetch on real hardware. No payment until it's verified on the real duck.

All numbers and footage on this page were measured in simulation (MuJoCo, September 2026) with the open MicroDuck model and its shipped walking and pick policies. The watch is real hardware built by InHandPlus; real-hardware validation is planned for December. Not affiliated with Hugging Face or Pollen Robotics.