Active Cross-Modal Visuo-Tactile Perception of Deformable Linear Objects

TL;DR

This work addresses the challenge of reconstructing the full geometry of deformable linear objects under occlusion by fusing foundation-model-based visual segmentation with autonomous tactile exploration. The authors propose a deterministic cross-modal pipeline that extracts a topology-aware visual representation via skeletonization, complements it with tactile data to recover occluded segments, and merges the data through endpoint-guided sorting followed by a B-spline reconstruction. Key contributions include topology-aware endpoint detection, autonomous tactile exploration for missing geometry, and a unified fusion that yields smooth, global cable models even with severe occlusions. The approach is validated on a real robot with an RGB-D camera and tactile sensing, demonstrating reliable reconstruction for single and multiple cables on inclined and horizontal planes, highlighting its potential for robust deformable-object manipulation in practical settings.

Abstract

This paper presents a novel cross-modal visuo-tactile perception framework for the 3D shape reconstruction of deformable linear objects (DLOs), with a specific focus on cables subject to severe visual occlusions. Unlike existing methods relying predominantly on vision, whose performance degrades under varying illumination, background clutter, or partial visibility, the proposed approach integrates foundation-model-based visual perception with adaptive tactile exploration. The visual pipeline exploits SAM for instance segmentation and Florence for semantic refinement, followed by skeletonization, endpoint detection, and point-cloud extraction. Occluded cable segments are autonomously identified and explored with a tactile sensor, which provides local point clouds that are merged with the visual data through Euclidean clustering and topology-preserving fusion. A B-spline interpolation driven by endpoint-guided point sorting yields a smooth and complete reconstruction of the cable shape. Experimental validation using a robotic manipulator equipped with an RGB-D camera and a tactile pad demonstrates that the proposed framework accurately reconstructs both simple and highly curved single or multiple cable configurations, even when large portions are occluded. These results highlight the potential of foundation-model-enhanced cross-modal perception for advancing robotic manipulation of deformable objects.

Figures (18)

• Figure 1: Block diagram of the proposed cross-modal visuo-tactile perception framework: visual perception is triggered by a natural language prompt for a state-of-the-art segmentation network; a clustering algorithm isolates different cables based on their colour; the segmented image is converted into point clouds and post-processed triggering the tactile exploration; finally a B-spline interpolation provides a model of the DLO.
• Figure 2: Experimental setup: robot arm with camera and tactile sensor.
• Figure 3: CS1 (i): camera RGB image (top-left), result of the semantic segmentation by Florence2/SAM2 (top-right), segmented cable (bottom-left) and support surface (bottom-right).
• Figure 4: CS1 (i): point clouds of the skeletonised cable (top-left) and of the inclined support surface (top-right). The skeleton is downsampled (bottom-left) and subsequently processed to replace the closer points with the midpoint, and then all points are projected onto the plane identified by RANSAC (bottom-right).
• Figure 5: CS1 (i): first sorted point cloud with endpoints in red (top-left), point cloud obtained by merging both visual and tactile point clouds (top-right), resulting point cloud of the new sorting step (bottom-left), interpolated point cloud (bottom-right).