Question about the velocity field of HeLiPR Aeva data used in Dynamic-ICP

[fengruhua]

Hi, thank you for the great work on Dynamic-ICP.

I have a question about the HeLiPR experiments, specifically regarding the Aeva per-point velocity field used by the method.

From my understanding, the Doppler term in Dynamic-ICP assumes that each point velocity is a raw relative radial Doppler measurement, i.e., for static points it should approximately satisfy:

s_i ≈ -u_i^T v_ego

However, when checking the public HeLiPR Aeva data, the velocity field seems to behave more like an absolute / ego-motion-compensated velocity rather than raw relative radial Doppler. For example, in a high-speed Bridge sequence, most static-background points have velocity values close to zero even when the INS/GT ego speed is around 10–25 m/s. Also, solving the raw-Doppler least-squares model from the point velocities does not recover the ego speed; it often gives near-zero or unstable estimates.

This seems consistent with the HeLiPR discussion here, where the Aeva velocity field is described as “absolute velocity” directly provided by the sensor:

minwoo0611/HeLiPR-File-Player#4

Could you clarify how the HeLiPR Aeva velocity was used in your experiments?

Specifically:

1. Did you directly use the public HeLiPR velocity field from the Aeva .bin files?
2. If yes, was it treated as raw relative radial Doppler, or as an absolute / ego-compensated velocity?
3. Did you apply any preprocessing or conversion to obtain relative radial Doppler velocities?
4. If the public HeLiPR velocity is absolute, does the Doppler ego-motion estimation part of Dynamic-ICP become disabled or effectively rely on the geometric ICP term and velocity-based dynamic filtering?

I am trying to reproduce the HeLiPR experiments and want to make sure I am using the velocity field in the same way as your implementation.

Thanks a lot!

Best regards