Unitree Go2 Quadruped State Estimation
Dataset This dataset for legged robot state estimation consists of recordings of the Unitree Go2 quadruped robot performing locomotion both indoors and outdoors. The aim is to evaluate proprioceptive state estimation using an IMU and leg odometry, both of which are recorded at 1000 Hz as
Dataset
This dataset for legged robot state estimation consists of recordings of the Unitree Go2 quadruped robot performing locomotion both indoors and outdoors. The aim is to evaluate proprioceptive state estimation using an IMU and leg odometry, both of which are recorded at 1000 Hz as bag files with ROS2 Humble. The robot performs two gaits: trotting and a bio-inspired adaptive gait which switches from walking to trotting based on the commanded velocity.
- The indoor experiments (L) are conducted at DFKI RIC in Bremen, Germany, with Vicon motion capture as ground truth. We provide 9 trajectories covering a total distance of 680 m.
- The outdoor experiments (F) are conducted on rocky terrain on Vulcano Island in Italy and contain accurate RTK/GNSS position data as ground truth. There are 18 trajectories, covering a total distance of 530 m.
Overview
An overview of the trajectories is provided in the table below. “L” stands for Lab (indoor) and “F” stands for Field Test (outdoor). For a quick look, you can visualise the plotted trajectories in the indoors_plots.zip and outdoors_plots.zip files.
Indoor | ||||
|---|---|---|---|---|
| Index | Gait | Terrain | Trajectory | Distance [m] |
| L1 | trotting | flat | random | 22.0 |
| L2 | trotting | stones | random | 31.1 |
| L3 | trotting | flat | turning counterclockwise | 51.4 |
| L4 | trotting | flat | turning clockwise | 64.7 |
| L5 | adaptive | flat | random | 26.4 |