Dataset Viewer
Auto-converted to Parquet Duplicate
instruction
stringclasses
5 values
robot_interpretation
stringclasses
5 values
robot_response
stringclasses
5 values
human_feedback
stringclasses
5 values
Can you bring me the cup? The one near the window.
Identify cup closest to window and deliver it
Moved to window area, picked up cup, delivered to user
Correct, but a bit slow
Wait here, I might need help in a minute.
Pause movement and remain attentive
Stopped movement and entered standby mode
Good, stayed responsive
No, not that one… the other box.
Previous object selection incorrect, re-evaluate
Released first box and selected alternative
Better, that’s the right one
I’m not sure, maybe just clean this area.
User expresses uncertainty but requests cleaning
Started light surface cleaning
Okay, that works
Stop for now.
Immediate task interruption
Ceased all actions safely
Good response

Human–Robot Interaction Notes (Manual)

This dataset contains manually written interaction notes describing how a humanoid agent responds to everyday human instructions.

The goal is to capture natural, imperfect, and human-like interactions instead of idealized or synthetic commands.


Motivation

While many datasets focus on clean and optimal commands, real-world human interaction is often ambiguous, emotional, and context-dependent.

This dataset was created to explore how humanoid agents can better understand natural human intent.


Data Collection Process

All entries were written manually by hand.

The interaction notes are based on:

  • Daily observations
  • Common human instructions
  • Realistic hesitation, uncertainty, and correction

No automation tools or scraping pipelines were used.


Contents

Each record includes:

  • Human instruction
  • Robot interpretation
  • Observed response
  • Human feedback

Use Cases

  • Human–robot interaction training
  • Natural language grounding
  • Behavior alignment for humanoid agents

Part of

Humanoid Network (HAN)


License

MIT

Downloads last month
5