Full six-degree-of-freedom rigid-motion modeling

Develop an extension of RVLoss, the self-supervised runoff-vote loss for LiDAR scene-flow estimation, that explicitly models full six-degree-of-freedom rigid motion, including rotational motion, which becomes increasingly important over longer temporal sequences.

Background

RVLoss generates cluster-wise rigid-flow pseudo-labels through a two-stage runoff-vote mechanism, but its current formulation represents motion using translations and does not explicitly account for rotation. The paper notes that this omission becomes more consequential when the temporal interval or sequence length increases, because rigid objects may undergo appreciable rotational motion.

The authors therefore identify handling full six-degree-of-freedom rigid motion—three translational and three rotational degrees of freedom—as unresolved future work. This problem concerns extending the current translational voting and loss framework rather than merely changing the backbone architecture.

References

Additionally, RVLoss does not explicitly model rotational motion, which becomes increasingly important over longer temporal sequences. Extending the framework to handle full 6-DoF rigid motion remains future work.

— RVLoss: Runoff Vote Loss for Self-Supervised LiDAR Scene Flow Estimation  (2608.18864 - Wang et al., 19 Aug 2026) in Section 'Conclusions', subsection 'Limitations and Future Work'

(iii) The current framework requires geometry (e.g., depth and estimated current and past states) as input; relaxing these requirements and enabling direct inference from RGB videos are also directions for future work.

— World Motion Models: Flexible Sequence Modeling of SE(3) Trajectories  (2610.01742 - Lei et al., 1 Oct 2026) in Section 6, Conclusion, paragraph “Limitations”