Dexhands

DEXTEROUS MANIPULATION

The world of dexterous hands.

Compare hardware. Discover applications. Share what works. An open reference and community for dexterous manipulation.

Dexterous Hands · Course contents
Learn / Evidence & Evaluation
5 min reading + activities · Research path

Make a hand experiment reproducible

A useful result describes what another person must repeat, including the failures.

Define placement success and record every attempt. Use the same setup and intervention rules when comparing results.Concept illustration · not a product specification or test result.
What you will learn

Produce a test record another person can challenge and repeat.

Decide the question before the trial

Choose one manipulation task, starting conditions and success rule. Report all attempts and explain excluded trials. Keep objects, controller, intervention policy and timing boundaries consistent when comparing systems.

Describe the complete system

Record the hand revision, external actuation, sensors, firmware, code commit, dependencies and configuration. For a build, include editable design files and the bill of materials where you have permission. Hardware, software and data may have separate licenses.

Use the right reference

NIST manipulation research and YCB protocols offer starting points for task design. RobotPerf measures robotics computing performance; its latency results do not rank hand dexterity. A draft standards work item is not an adopted standard.

Publish enough to challenge the claim

Use the experiment template to record the procedure, successes, failures, resets, human assistance, logs and limitations. Link permitted artifacts to their original repository. A video can illustrate the result; it does not replace the procedure or trial record.

Try it yourself

EVIDENCE LAB · invented teaching data

The denominator changes the story.

✓ Unassisted success+ Assisted success× Failure

Assisted successes are a subset of all successes here. “Unassisted” alone does not establish autonomy: the control mode and intervention policy still need to be documented. These counts describe a sample, not a guaranteed future success rate.

Your practical task

  1. Choose one task, object and starting condition.
  2. Define success, failed attempts and allowed interventions before testing.
  3. Open the experiment template and record all outcomes and missing evidence.

What to produce: A completed experiment template, with unknowns explicitly marked.

Check your understanding

Which record is most useful for reproducing a manipulation result?

Original sources & next steps

NIST: manipulation measurement research ↗YCB benchmarking project ↗OSHWA open hardware definition ↗

Original Dexhands teaching material. Interactive models and added references reviewed September 23, 2026; the introductory reading retains its September 22 review. Research links are not endorsements or evidence of hardware compatibility.

Start a discussion

Published posts are public. First contributions are reviewed. Only share material you have permission to disclose.

Report a post