Research Analyzer
← Back ICRA 2026

DisFlow: Scene Flow from Distance Field for Object Pose, Velocity Tracking, and Surface Reconstruction

Lan Wu, Sheila Sutjipto, Jennifer Wakulicz, Teresa A. Vidal-Calleja

PDF

AI summary

Key figure (auto-extracted from paper)
DisFlow enables real-time, uncertainty-aware 6DoF pose and velocity tracking alongside dense surface reconstruction by computing scene flow directly from a Gaussian Process distance field.
Scene flow Gaussian Process Implicit Surfaces Dynamic object tracking Real-time reconstruction 6DoF pose estimation Probabilistic mapping

Problem

Existing methods struggle to jointly estimate accurate object motion and dense surface geometry in real-time, often lacking uncertainty estimates or failing with dynamic objects.

Approach

The method represents objects as Gaussian Process Implicit Surfaces to generate a continuous signed distance field, then computes scene flow from this field to incrementally register point clouds via closed-form optimization in an object-centric frame.

Key results

  • Achieves 92-95% ADD-AUC and sub-centimeter translation error on Fast-YCB dataset
  • Delivers real-time linear and angular velocity estimation with uncertainty bounds
  • Produces dense, temporally consistent surface reconstructions without dynamic point filtering
  • Outperforms tracking-only and reconstruction-only baselines in accuracy and robustness

Why it matters

Provides a unified, uncertainty-aware perception pipeline essential for safe robotic manipulation, grasping, and human-robot interaction with moving objects.

Abstract

We present DisFlow, a novel framework for online scene flow estimation from distance field that enables 6DoF dy- namic object pose estimation, motion tracking, and surface recon- struction. The scene is represented by Gaussian Process Implicit Surfaces (GPIS), with surface normals serving as derivative constraints, enabling accurate signed distance computations near the surface and gradient queries with uncertainty. With this representation as a foundation, we compute a scene flow from the distance field that describes how surface points are transported over time in consecutive frames. Through our flow, we can estimate an object’s pose and motion by incrementally registering a new observed point cloud via an elegant closed- form optimisation. Unlike prior methods that operate in the camera or world frame, our approach performs probabilistic fusion directly in the object frame, where the object remains geometrically consistent over time. The tight coupling of the DisFlow method in space and time yields dense geometry, surface normals, object pose trajectories, velocities, and uncer- tainty, all at real-time rates. We evaluate DisFlow on dynamic object sequences and demonstrate that it achieves accurate pose and motion tracking while simultaneously reconstructing high- quality object surfaces.

Index terms

Mapping RGB-D Perception

Related papers