Datasets

The dataset was collected across multiple days on different robots in a number of indoor environments. The dataset has been divided into multiple groups based on the environment in which it was collected.

Description

Each dataset consists of sensor data from three different modalities:

1. range or distance data obtained using DW1000-based Bitcraze Loco Positioning ultrawideband (UWB) radios,
2. body-referenced linear acceleration and angular velocities from an inertial measurement unit (IMU), and
3. visual inertial odometry (VIO) from Intel Realsense T265/T261 tracking camera.

Ground truth

Each dataset has the ground truth pose of the robot from a Vicon motion capture system for evaluation. In environments where a motion capture system is unavailable, ground-truth position information from a Leica Total station is provided.

Format

The sensor data from individual experiments is available in two formats: as rosbags and as csv files. When using rosbags, data from IMU, VIO and ground truth can be (de)serialized using standard ROS message definitions. Range data can be (de)serialized using the custom message type: measurement_msgs/Range. The ROS package for this custom message can be found here: link. The message fields are as below:

std_msgs/Header header
string mobile        # name of tag
string[] anchors     # name of anchor
float32[] data       # measured distance
float32[] covariance # measurement covariance

The position of tag in body frame and the position of the anchor in world frame can be obtained from the config files provided with each dataset.

Calibration

A big part of achieving accurate localization involves calibrating the different sensors and their spatial and temporal offsets. We provide the spatial offsets of the different sensors with respect to the robot center in each case.

The location of the anchors, body-referenced spatial offsets, and noise parameters are provided in the robot_config.yaml file associated with each dataset.