Performances of AprilTag and Intel RealSense Tracking Camera T265 on 4-Wheeled Robot

Fig. 1: Intel RealSense Tracking Camera T265 attached to the robot

By knowing the fact that detecting and publishing the AprilTag coordinates are time-consuming, attaching Intel RealSense Tracking Camera T265 to compare and analyze these two performances, with TF (transformations) system. If the tracking camera's performances in terms of time turn out to be better, then we will use the AprilTag to acquire the position of the robot in the world frame at first, then we will be relying on the tracking camera to track the position of the robot. The robot's location will be captured through the static global Xbox camera, by capturing the AprilTag attached to the robot, which we already set up beforehand.

Fig. 2: Rviz visualization. Fisheye works well, and the position of the camera frame correctly depicted

Visualization of the datasets is challenging due to the offset distance between the center of the AprilTag and the Intel Tracking Camera T265 (Fig. 2). Due to this offset, the visualization of the 3D plot does not look intuitive -the the flow of trajectories are same, but there always exists a distance gap.

Video 1: Rugged Terrain, with Random Robot Movement

Data Collected as follows:

Smooth Terrain, with Straight Robot Movement
Rugged Terrain, with Straight Robot Movement
Smooth Terrain, with Random Robot Movement
Rugged Terrain, with Random Robot Movement

Fig. 3: Trajectory analysis for the AprilTag (in red color) and the Intel Tracking Camera T265 (in green color) with the environment of unstructured terrain, with random movement of the robot. The gap between coordinates is the distance offset between the center positions.

This last experiment has the most difference between the tracking camera and the AprilTag. Unlike the first three 3D trajectory, once I compare the actual trajectory of the robot (Video 1) to the 3D scatter plot (Fig. 3), there are some missing coordinates in the trajectory detected by the AprilTag. Once you take a look at the x-axis from -120 to -140, some of the red coordinates are missing -these missing coordinates does not make sense because the robot cannot just jump from one position to another. However, in case of the camera, it precisely and continuously detected the trajectory where the robot was circulating around. Thus, in can be inferred that there are some delays exists in the AprilTag; in other words, the performance of the AprilTag is relatively slower compared to the tracking camera.