Research Analyzer
← Back ICRA 2026

MIND - Multi-Feature Implicit Neural Descriptors for Robotic Surface Processing of 3D Objects with Variations in Geometry

Anish Pratheepkumar, Christian Hartl-Nesic, Markus Ikeda, Fabian Widmoser, Andreas Pichler, Markus Vincze

PDF

AI summary

Key figure (auto-extracted from paper)
MIND enables one-shot transfer of robotic surface processing trajectories across geometrically diverse objects without CAD models, outperforming existing dense correspondence methods.
Dense correspondence Implicit neural descriptors Robotic surface processing CAD-free automation Geometric variation One-shot transfer

Problem

Robotic surface processing struggles to generalize across objects with high geometric variation in high-mix manufacturing, as existing methods rely on CAD models or require tedious reprogramming.

Approach

The authors introduce MIND, a neural implicit descriptor that fuses local geometry, normals, signed distance functions, and sparse keypoints to establish dense 3D correspondences across category-level objects, allowing direct trajectory transfer from raw point clouds.

Key results

  • Novel multi-feature implicit neural descriptor architecture
  • CAD-free one-shot trajectory and process knowledge transfer across geometric variations
  • Quantitative and qualitative outperformance of state-of-the-art dense correspondence methods
  • Real-world validation on robotic surface processing of geometry-varying basin molds

Why it matters

Provides a scalable, CAD-free solution for flexible robotic automation in high-mix manufacturing environments.

Abstract

Therecentshiftfrommassproductiontomassperson- alization leads to a production environment in which workpieces have a high degree of geometric variations. The robotic process automation in such high-mix low-volume environments poses sig- nificant challenges since predetermined robot programs are not viable anymore. In this letter, we consider the automation of surface processing for category-level objects with significant variations in geometry by operating on point clouds without relying on CAD models. To achieve this, we present a novel multi-feature implicit neural descriptor (MIND) representation which leverages dense correspondence to generalize across diverse objects, enabling a one-shot transfer of process trajectories and associated process knowledge. The quantitative and qualitative evaluation shows that MIND outperforms other state-of-the-art dense correspondence approaches. A real-world application case study of robotic surface processing on geometry-varying basin molds validates the efficacy of the proposed approach.

Index terms

Computer Vision for Automation Industrial Robots Representation Learning

Related papers