Reproduction 002 · OpenSee all studies

Pipetting robot clip study

A claim of 69% success over three tasks travelled with a twenty-second clip that shows one of them, once. Here is what the clip proves and what our runs show.

What the pipetting clip shows in twenty seconds.

A fixed camera on a lab bench and two five-finger hands; only one moves, so the task as filmed is one-handed. No arm, torso or head is visible.

2 s
The hand hovers above the flask
8 s
The pipette is lifted clear
13 s
The tip is in the beaker
18 s
Released across the rim
Four moments from the clip: hovering, lifting the pipette, the tip in the beaker, the release

Six things the pipetting clip does not contain.

None of this makes the claim false. It makes it unverified: a clean single take is a demonstration, and 69% is a measurement.

The cited preprint is not linked, and two searches, by name and by the three tasks, found nothing matching.

Evidence in the clip1 take
Touch signal from the fingertipsMissing
Run without touch, as a baselineMissing
A slip, push or recoveryMissing
Number of trials behind 69%Missing
Playback speed (project videos run at 5x)Unstated
Primary sourceNot found

Three blocks to reproducing the pipetting method.

  • 1 · No touch channel

    Our G1 and Inspire model has no per-fingertip touch signal, so a policy built on one can be imitated, not evaluated.

  • 2 · No liquid in MuJoCo

    Drawing up and dispensing have no match. Grasp, lift, carry and release can be measured, and we will not score them as the harder task.

  • 3 · Resolved: no GPU on this host

    The work moved to rented GPUs, checked with nvidia-smi inside the containers; π0 inference runs on an A10.

The robot hand holding the pipette above the flask

What has run on the AutoBio pipetting scene.

A published MuJoCo benchmark with a pipetting task already exists, AutoBio (ICLR 2026), so we ran that, unmodified, instead of scoring a private copy.

  • The loop runs, the task does not

    All fourteen joints move for thirty cycles, yet the free body moved 0.912 mm and the tip stayed 74.8 cm or more from the tube.

  • Our own classifier was wrong first

    It first counted the tube resting in its rack as contact; both of its errors flattered the result.

AutoBio pipette runπ0 base
Robot: dual UR5e, DexHand and 2F-85Unmodified
Simulator: MuJoCo 3.3.0Pinned
Harness check, bit for bit on rerun100 / 100
Loop: 30 cycles, peak 1.02 rad/sRuns
Contacts with objects0
Task doneNo
Benchmark install as pinnedBroken

Where the claim stands

We still have no number on the 69% claim, and we will not dress nearby results up as an answer to it.