Robot Metrics
As of 08.05.26, only Furhat MS has been released with proper metrics collection.
Furhat MS collects raw data (event agnostic states) approximately every 1 second (be default) when an action script is being executed. The data contains a timestamp and a list of users (humans) detected by Furhat (with some info like coordinates, distance, etc.), collected via RealTime API. These raw data are then processed into usable “robot metrics”.
Robot metrics calculated by Furhat MS:
| Metric slug | Type | Meaning |
|---|---|---|
face_detection | float or None | Ratio of time a face was detected over the effective duration of an action script |
attention_to_robot | float or None | Ratio of time the closest detected face was looking at the robot over the effective duration of an action script |
sitting | float or None | Ratio of time the closest person was sitting (assumed true if the distance is <= 1 meter) over the duration bla-bla |
standing | float or None | Same as sitting, but assumed true if distance is > 1 meter |
proximity | float or None | Average distance to the closest detected person; not a ratio |
happiness | float or None | Average happiness measure based on emotion recognition; calculated as average confidence of happy emotion |
In the table, an “effective duration of an action script” means the time between the action script (1) started being processed (in queue processing coroutine) and (2) updated status to stopped or completed. Note that this happens strictly within the MS and is not connected to the offline backend.
In the table, ratio values are bound from 0 to 1, inclusively. proximity is not bound.