About
The dataset comprises ~60 minutes of data split approximately equally across four different map configurations, each with three runs. For each map configuration, we obtain a map of the arena by using the ROS hector mapping package with the Toyota HSR’s onboard Hokuyo UST-20LX laser range sensor. To provide additional variation, our dataset comprises two people, with the map configurations split equally between them. We use a high-frequency (100Hz) Vicon motion capture system to provide ground-truth tracking of humans, the robot, external camera positions, and goal locations.
Data Sources:
- ASUS Xtion Pro Live – an onboard (robot-mounted) RGB-D camera
- Intel Realsense D435 – a static/external RGB-D camera
- Vicon Motion Capture System (100Hz)
Arena Setup
We recorded the dataset within a designated arena (7.1x4.2m) marked out within a large laboratory and under a Vicon motion capture system. The dataset comprises four different configurations of obstacles/goal locations. For each map configuration, we collected three ~5-minute runs of human behaviour. During each run, a person walks repeatedly between both goal locations and entrance/exit locations for the arena. Meanwhile, a human operator moves the Toyota Human Support Robot around the arena, attempting to maintain vision of the person with the head-mounted RGB-D camera.
Environment Mapping
For each of the four obstacle configurations in the dataset, we provide a map generated using the HSR’s onboard laser range sensor (Hokuyo UST-20LX) with the ROS package hector mapping.
Motion Capture Setup
To acquire the dataset, we calibrated the motion capture arrangement (18 Vicon Vero 2.2 cameras 2.2MP with 850nm IR emitters) cameras to achieve a sub-millimetre average residual tracking.
For person tracking, we mount reflective IR markers onto helmets with unique 3D arrangements. To obtain accurate goal locations we use similarly tracked objects placed at each obstacle/goal location. For camera sensors, we use custom 3D-printed mounts with unique marker arrangements.
ROS Topics
Each rosbag contains the following list of topics:
| Topic | Frequency [Hz] | Description |
|---|---|---|
/vicon/world |
100 | Pose of the vicon origin |
/tf |
30 | TF transforms (e.g. robot and camera frames) |
/tf/static |
30 | Static TF transforms (e.g. fixing camera to world frame) |
/vicon/bottle_4_markers/bottle_4_markers |
100 | Pose of the bottle/bin marker |
/vicon/d435_4\_markers/d435_4_markers |
100 | Pose of the static camera marker origin |
/vicon/entrance_3_markers/entrance_3_markers |
100 | Scene entrance/exit marker |
/vicon/exit_3_markers/exit_3_markers |
100 | Second scene entrance/exit marker |
/vicon/person_1/person_1 |
100 | Pose of the helmet/person’s head |
/vicon/plant_3_markers/plant_3_markers |
100 | Pose of the plant marker |
/vicon/table_4_markers/table_4_markers |
100 | Pose of the table marker |
/camera/depth/image_rect_raw/compressed |
30 | Static camera depth images |
/camera/color/image_raw/compressed |
30 | Static camera rgb images |
/hsrb/head_rgbd_sensor/depth_registered/image_rect_raw/compressed |
30 | Robot camera depth images |
/hsrb/head_rgbd_sensor/rgb/image_rect_color/compressed |
30 | Robot camera rgb images |