RH-IoT Dataset

University of Hertfordshire Robot House IoT Dataset for Room-Level Human Presence Detection

RH-IoT is an ambient sensor dataset collected in the University of Hertfordshire Robot House. Robot House is a smart home environment used for research in human-robot interaction and assistive technologies.

The dataset captures natural interactions between a resident and the home environment using non-intrusive ambient sensors. Data were recorded under natural conditions rather than scripted activities.

Dataset Characteristics

Property	Value
Total timestamps (samples)	63,438
Total sensors recorded	68
Binary sensors	33
Numeric sensors	35
Annotation columns	`activity`, `location`
Timestamp format	ISO 8601
Date range	2025-01-14 to 2025-01-20

File Structure

The dataset is released as two files:

`RH_IoT_Dataset.csv` — Main data file

139 columns structured as follows:

timestamp — ISO 8601 datetime string, one row per recording interval
Sensor columns — grouped per sensor: SensorName_STATUS followed by SensorName_VALUE, repeated for all 68 sensors
location — room-level location label
activity — activity label

Each sensor contributes two columns:

Attribute	Description
`SensorName_STATUS`	Binary sensor state, pre-encoded as integer 0 or 1
`SensorName_VALUE`	Numeric measurement. Binary sensors mirror STATUS (0/1). Numeric sensors are min-max normalised to [0, 1].

Spatial coordinates (XCOORD, YCOORD) are provided separately in sensor_metadata.csv.

`sensor_metadata.csv` — Sensor positions

68 rows, one per sensor. Columns:

Column	Description
`sensor_name`	Matches the sensor name prefix used in the main file
`xcoord`	Sensor x-position on the house layout
`ycoord`	Sensor y-position on the house layout

Researchers requiring spatial information can join this file to the main dataset on sensor_name. Note: the watchingtv sensor entry has same coordinates as TV.

STATUS Encoding

All _STATUS columns are pre-encoded as integers:

Raw value	Encoded	Notes
`On`, `Open`, `Present`	1	Standard active state
`Off`, `Closed`, `Absent`	0	Standard inactive state
`Free`	1	Pressure/seat sensors: no weight detected
`Occupied`	0	Pressure/seat sensors: weight detected
`Closed`	1	Toilet Lid only — closed lid is the default/normal state
`Open`	0	Toilet Lid only

Sampling Rate

1-second intervals — majority of recordings
2-second intervals — secondary mode
Occasional larger gaps reflect natural recording downtime

No artificial resampling was applied.

Annotation

`activity`

Labels are compact single-word strings (no spaces):

Label	Description
`cooking`	Cooking at the stove or oven
`dishes`	Washing dishes
`havingmeal`	Eating a meal
`havingtea`	Drinking tea
`mealprep`	Preparing a meal
`prayers`	Prayer activity
`prephotdrink`	Preparing a hot drink
`sitting`	Sitting (general)
`standing`	Standing (general)
`toileting`	Toilet use
`walking`	Walking or moving between rooms
`watchingtv`	Watching television
`other`	Activities with fewer than 300 instances, grouped to reduce class imbalance(hygiene: 121 instances, resting: 89, and preparing cold drinks: 85 instances)

`location`

Seven room-level locations:

Label	Description
`bathroom`	Bathroom
`bedroom`	Bedroom
`corridor`	Corridor / hallway
`diningarea`	Dining area
`kitchen`	Kitchen
`sofaarea`	Sofa / living area
`staircase`	Staircase (leading to upstairs area)

Labels were created through manual review of synchronised video recordings. The videos were used only to confirm ground truth and are not included in the released dataset.

Known Limitations

Data were collected from a single resident.
Non-uniform daily recording durations (3–5 hours per day typical).
Natural downtime gaps between recording sessions.
Class imbalance in rare activities, addressed by grouping into other.

Intended Use

The dataset can support research in:

Room-level human presence detection
Assistive robotics and human-robot interaction
Hospital-at-home and ambient assisted living
Smart home context modelling
Time-series machine learning

Access

The dataset can be downloaded from the Robot House dataset repository.

For access requests please contact: robothouse@herts.ac.uk

Citation

If you use this dataset in your research, please cite the associated paper:

Sehrish Rafique, Patrick Holthaus, Gu Fang, and Farshid Amirabdollahian. RH-IoT-1: A Real-World Smart-Home Dataset for Room-Level Human Presence Detection, International Conference on Social Robotics (ICSR 2026).

@inproceedings{rafique2026rh,
  title={RH-IoT-1: A Real-World Smart-Home Dataset for Room-Level Human Presence Detection},
  author={Rafique, Sehrish and Holthaus, Patrick and Fang, Gu and Amirabdollahian, Farshid},
  booktitle={Proceedings of the 18th International Conference on Social Robotics (ICSR 2026)},
  series={Lecture Notes in Computer Science},
  publisher={Springer},
  address={London, UK},
  year={2026}
}