AGHRI: agricultural human-robot interaction dataset

AGHRI is a multi-sensor dataset of people moving and working around an agricultural robot. It was recorded to develop and benchmark the human detection and tracking that robots need to work safely alongside farm workers.

Scope 65 sequences, about 27 minutes of recordings, about 70 GB, 10 participants
Sensors ZED RGB-D camera, three fisheye cameras and a Livox Mid-360 LiDAR
Annotations 154,002 2D bounding boxes and 27,311 3D bounding boxes
Recorded University of Lincoln Riseholme campus, October to November 2024
Licence CC BY 4.0
Download doi.org/10.24385/lincoln.32982638

Code for working with AGHRI

  • AGHRI-dataset-benchmark: baselines for 2D and 3D human detection and tracking, camera and LiDAR fusion, and person re-identification (Apache-2.0).
  • AGHRI-dataset-tools: converters to KITTI, YOLO, COCO, MMDetection3D and MOT formats, and ROS 2 bag generation (MIT).

The dataset is used in the camera and LiDAR late fusion practical at the ROS in the Field workshop, ROSCon UK 2026.