[Question] body linear velocity acquisition and odometry logic for G1-EDU

[blyu413]

Hi Unitree Team,

I am currently working with the Unitree G1-EDU and am trying to accurately obtain the linear velocity of the robot's body. I have a few questions regarding the velocity signals and the underlying estimation logic:

1. Regarding the reference point of rt/odommodestate According to the documentation (Odometer Service Interface) , the velocity can be obtained by subscribing to rt/odommodestate. The documentation describes this as the "velocities of the robot's base center."

• Question: Could you clarify which exact point this "base center" refers to? Specifically, does it correspond to the origin of the torso coordinate system (e.g., the pelvis or base_link frame) as defined in the MuJoCo model?

2. Odometry calculation logic and suspended behavior During my experiments, I noticed a specific behavior:

• When the robot is suspended by a crane and pushed forward, the velocity signal does not change.
• However, when the robot is walking on the ground, the velocity signal reflects the movement correctly.
• Question: This suggests the odometer might be relying heavily on leg kinematics (contact sensors/joint encoders). Could you provide more details on how the velocity and displacement in the odometry service are calculated? Is it a fusion of IMU and leg kinematics, and why does it fail to register movement when off the ground?

3. Alternative velocity signals (Sensor Fusion) Does the Unitree G1-EDU provide any other velocity topics or estimators beyond the standard odometry?

• Question: For instance, is there a built-in state estimator that provides fused velocity data using the Livox Mid360 and IMU out of the box?