Sensor Setup

- 1 min

Sensor Setup

We argue that while smooth motion, clear images, and dense point clouds obviously facilitate algorithm processing, they may not be sufficient for practical applications. Therefore, we select cost-effective and widely used sensors to replicate real-world robotic application conditions as closely as possible.

figure

  1. LiDAR: We utilize the Velodyne VLP-16 LiDAR to capture 3D point clouds of the surrounding environment at 10 Hz, collecting over 300,000 points per second. The sensor features a 360° horizontal field of view (FoV) and a 30° vertical FoV, with a scanning range over 100 m. To ensure efficient data transmission, we connect the LiDAR to the host via an Ethernet port.

  2. Camera: The Intel RealSense D455 is a global shutter RGB-D camera capable of capturing time-synchronized color and depth images at a resolution of 720×480 pixels at 10 Hz, whose optimal depth range extends from 0.6 m to 6 m from the image plane, with an error margin of less than 2% within a 4 m range.

  3. IMU: To ensure higher accuracy in IMU data acquisition, we select not to use the internal IMU (Bosch BMI055) of the RealSense D455. Instead, we integrate the Xsens MTi 300 to capture 9-axis measurement data at 200 Hz. The accelerometer exhibits a bias stability of 15 \(\mu g\) and a noise density of 60 \(\mu g/\sqrt{Hz}\), the gyroscope achieves a bias stability of 10 \(°/s\) with a noise density of 0.003 \(°/s\), and the magnetometer provides a resolution of 0.25 \(mG\).

  4. GNSS: The M68UGI-G GNSS receiver is utilized to acquire raw positioning data, including GPS satellite count, timestamps, signal quality, and other relevant information at 5 Hz. By leveraging RTK technology, high-precision robot positions are estimated as ground truth for the trajectory, achieving a horizontal accuracy of \(\pm(8mm+1ppm)\) and a vertical accuracy of \(\pm(15mm+1ppm)\).

comments powered by Disqus