How to look beyond a great demonstration
A short clip is a starting point for questions, not a complete performance test.
Report denominators and interventions alongside successful outcomes.
Describe the conditions
Record the object, surface, starting pose and task. Does the system work when any of those change? Identify whether the video is edited or accelerated when that is disclosed.
Look for repeatability
Ask for the number of attempts and the definition of success. Note failures, resets, human help and recovery. A single successful run does not establish a success rate.
Make a useful observation
Separate “the video shows” from “the system can always do.” Link the original source and describe what is missing. Your careful question is useful even if you do not own hardware.
Try it yourself
EVIDENCE LAB · invented teaching data
The denominator changes the story.
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
- Change the successes and assisted-success counts in the teaching ledger.
- Compare all successful outcomes with unassisted successful outcomes.
- Write down the task, attempt boundary and intervention rule you would use.
What to produce: A one-sentence claim with its denominator, assistance rule and limitation.
Check your understanding
Original sources & next steps
Open Robotics: evidence-based community discussion ↗NIST: manipulation performance measurement ↗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.