Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-Object RANSAC: Efficient Plane Clustering Method in a Clutter

Seunghyeon Lim, Youngjae Yoo, Jun Ki Lee, Byoung-Tak Zhang

TL;DR

MO-RANSAC presents a clutter-focused plane clustering method for RGB-D data that combines Deep Plane Clustering (DPC) with a post-processing merge to achieve plane instance segmentation in challenging scenes. DPC uses self-supervised pseudo-labels from RANSAC, along with voting layers, to create subplane clusters, which are refined through a merge step to form final planes. The approach achieves superior plane segmentation on OCID and OSD datasets and translates to tangible gains in real-world suction grasping with a UR5 robot, outperforming several RANSAC baselines and vision-based grasping methods. These results highlight MO-RANSAC’s potential for robust scene understanding and manipulation in cluttered environments, enabling more reliable robotic perception and control.

Abstract

In this paper, we propose a novel method for plane clustering specialized in cluttered scenes using an RGB-D camera and validate its effectiveness through robot grasping experiments. Unlike existing methods, which focus on large-scale indoor structures, our approach -- Multi-Object RANSAC emphasizes cluttered environments that contain a wide range of objects with different scales. It enhances plane segmentation by generating subplanes in Deep Plane Clustering (DPC) module, which are then merged with the final planes by post-processing. DPC rearranges the point cloud by voting layers to make subplane clusters, trained in a self-supervised manner using pseudo-labels generated from RANSAC. Multi-Object RANSAC demonstrates superior plane instance segmentation performances over other recent RANSAC applications. We conducted an experiment on robot suction-based grasping, comparing our method with vision-based grasping network and RANSAC applications. The results from this real-world scenario showed its remarkable performance surpassing the baseline methods, highlighting its potential for advanced scene understanding and manipulation.

Multi-Object RANSAC: Efficient Plane Clustering Method in a Clutter

TL;DR

MO-RANSAC presents a clutter-focused plane clustering method for RGB-D data that combines Deep Plane Clustering (DPC) with a post-processing merge to achieve plane instance segmentation in challenging scenes. DPC uses self-supervised pseudo-labels from RANSAC, along with voting layers, to create subplane clusters, which are refined through a merge step to form final planes. The approach achieves superior plane segmentation on OCID and OSD datasets and translates to tangible gains in real-world suction grasping with a UR5 robot, outperforming several RANSAC baselines and vision-based grasping methods. These results highlight MO-RANSAC’s potential for robust scene understanding and manipulation in cluttered environments, enabling more reliable robotic perception and control.

Abstract

In this paper, we propose a novel method for plane clustering specialized in cluttered scenes using an RGB-D camera and validate its effectiveness through robot grasping experiments. Unlike existing methods, which focus on large-scale indoor structures, our approach -- Multi-Object RANSAC emphasizes cluttered environments that contain a wide range of objects with different scales. It enhances plane segmentation by generating subplanes in Deep Plane Clustering (DPC) module, which are then merged with the final planes by post-processing. DPC rearranges the point cloud by voting layers to make subplane clusters, trained in a self-supervised manner using pseudo-labels generated from RANSAC. Multi-Object RANSAC demonstrates superior plane instance segmentation performances over other recent RANSAC applications. We conducted an experiment on robot suction-based grasping, comparing our method with vision-based grasping network and RANSAC applications. The results from this real-world scenario showed its remarkable performance surpassing the baseline methods, highlighting its potential for advanced scene understanding and manipulation.
Paper Structure (14 sections, 3 equations, 6 figures, 4 tables, 2 algorithms)

This paper contains 14 sections, 3 equations, 6 figures, 4 tables, 2 algorithms.

Table of Contents

  1. INTRODUCTION
  2. Related Works
  3. Plane Clustering
  4. Suction Grasp in a Dense Clutter
  5. Methods
  6. Overview
  7. Deep Plane Clustering
  8. Post-processing Module
  9. Experiments
  10. Experiments Setup
  11. Qualitative Results
  12. Plane Instance Segmentation Result
  13. Suction Grasp Performance in the Real World
  14. CONCLUSIONS

Figures (6)

  • Figure 1: A plane clustering result of Multi-Object RANSAC (MO-RANSAC, our method).
  • Figure 2: Overall framework of Multi-Object RANSAC. Deep Plane Clustering (DPC) takes a 9D point cloud and produces votes in the form of $\{\Delta x\}_{i=1}^N$ s.t. $\Delta x \in \mathbb{R}^3$. Meanwhile, DPC initially samples $K$ points (triangle-shaped points), to individually represent clusters after voting (large circle-shaped points, S). Each point is then associated with the nearest sampled point, resulting in the creation of $K$ subplane clusters. These subplane clusters undergo further refinement through post-processing following DPC.
  • Figure 3: Segmentation results compared to plane clustering baselines: red circles represent successful segmentation outcomes for small objects, while blue circles denote segmentation failures for nearby objects.
  • Figure 4: Ablation studies of MO-RANSAC and the ground truth.
  • Figure 5: Real-world examples.
  • ...and 1 more figures