Robot Cues GitHub
Human–robot interaction

When the robot
gives the cue.

Does it matter whether a cue comes from a screen or a robot? We followed 31 older adults through seven cognitive tasks, tracing what they saw, when they responded, and when they held back.

STUDY HARDWARE
Front view of the Booster K1 humanoid robot
iPad + Booster K1View in 3D
31older adults
ages 60 and up
07tasks in a
fixed sequence
02delivery conditions
tablet or tablet + robot

The protocol

Inside the seven tasks.

Everyone followed the same sequence. The iPad ran every task. In the robot condition, K1 spoke between tasks and provided movement cues during hand shape and direction matching.

Cue studio

Follow one cue.

In direction matching, tap when the movement and spoken word agree. If they conflict or the arm stays still, let the five-second window pass. Try both delivery conditions and inspect your event record.

Booster K13D model
Booster K1 model preview
Drag to rotate the model
Direction matchingSimulation
Cue source
iPad response

Ready to run a trial.

Event record

From cue to response.

The field app tied task order, cue timing, taps and timeouts to the same session. The robot bridge also reported movement readiness. Run a trial above to generate a browser-only example.

TRIAL EVENTSRELATIVE TIME
  1. Waiting for the first cue…
Inside a field session package

The iPad saved a session record and task-specific files. In the field-v4 workflow, a paired Mac checked file size, SHA-256 hashes and media readability. Explore the structure below.

Field findings

When holding back counts.

Older adults were randomly assigned to two groups: fifteen used the iPad alone and sixteen used it with K1. All 31 reached the end. Looking at individual response types reveals patterns that a single score would hide.

TabletTablet + robot
See outcomes for all seven tasks

Each row uses its own response unit. Eligible participants are listed in tablet / robot order. Rates describe these groups and do not form a clinical score.

Task / measure Tablet Tablet + robot Eligible people

Drawing allowed corrections and the tutorials differed between conditions. Hand-shape counts use 305 recorded windows. Delayed recognition requires standard three-target encoding.

Download aggregate table ↓