A1 SLAM: Quadruped SLAM using the A1’s onboard sensors
A1 SLAM is a rospackage that brings real-time SLAM capabilities utilizing factor graph optimization to Unitree’s A1 quadruped.
Note
This project is under active development.
Why use A1 SLAM?
Performance
A1 SLAM yields SLAM results competitive to the most prominent SLAM algorithms. Below are benchmarked Absolute Pose Error (APE) results compared to out-of-the-box Cartographer, one of the most widely-used 2D SLAM libraries available. The baseline was obtained using motion capture data.
Algorithm |
Trajectory 1 |
Trajectory 2* |
Trajectory 3* |
Trajectory 4 |
|---|---|---|---|---|
Cartographer |
0.04049 |
0.96218 |
1.39692 |
0.13435 |
A1 SLAM |
0.04681 |
0.11836 |
0.14518 |
0.04347 |
*Trajectories were recorded with noticeably more aggressive rotation and/or translation.
Convenience
- A1 SLAM is intentionally designed around using the sensors that are already onboard the A1 quadruped. This means:
Static transforms between sensor frames and the base link are already established
Algorithms tested on the onboard sensors for obtaining good performance
Documentation is provided for obtaining onboard sensor data (that may not be clearly documented elsewhere)