DiTer++ Description

Robot and Sensors

We utilize a quadrupedal robot (Unitree Go1 & Go2) as a base platform to collect the dataset and set up the compact sensor configuration.

Using a top-mounted intel NUC board, each sensor data was obtained asynchronously.

Sensor configuration

To capture visual measurements with different wavelengths, the RGB and thermal cameras are placed to form a forward-looking sensor system.
The terrain-looking sensor system uses an RGB-D camera and LiDAR to obtain sufficient ground information.
In addition, LiDAR enables the recognition of the surrounding environment regardless of angle.
Finally, IMU, and built-in state publisher are responsible for the robot's Navigation.

Frame configuration

Kinematic frame of each robot are provided based on the URDF information provided by Unitree
Each coarse extrinsic parameter are provided using CAD model. The following parameters are utilized as an initial guess during extrinsic calibration step.
We utilized identical calibration step used from our previous work, single-session dataset DiTer.

Data Format

We provide our datasets in rosbag format acquired in ROS noetic.
Each perceptual sensor and motion sensor are represented in this table.

Specification

Robot State

Each state of Robot are published in a single custom message format which includes robot's low-state information. Joint-state, built-in IMU, foot contact are included in the message.
Custom ROS message package is necessarily required to recognize and subscribe the following state. Required package is uploaded at our Github link.

Thermal Data

For those who extensively plans to utilize thermal image, thermal image are provided in 14-bit format.

Example Data from LAWN-NIGHT sequence

With the comprehensive multi-modal setup of our agents, visualization below includes :
- Inertial / Kinematic measurement
- Range measurement
- Thermal / RGB image
- Geometric / Photometric terrain measurement

Page updated

Google Sites

Report abuse