TransGraspNet: Physically and Geometrically Consistent Manipulation of Transparent Labware

Noosaga Placements

• Robot Manipulationsubfield95%
  The paper focuses on robotic grasping and manipulation of transparent labware, with grasp planning and execution as core contributions.
  TransGraspNet, a geometry–physics consistent framework for transparent object manipulationphysics-aware grasp refinement mechanism that explicitly incorporates task-level stability constraints for safe execution
• Learning-Based Manipulationframework90%
  The paper uses learning-based manipulation techniques, including deep learning for grasp candidate generation and refinement.
  generate initial 6D candidates using the official GraspNet-1Billion modelapply our geometric-physics constrained refinement as postprocessing
• Robot Perception And Slamsubfield80%
  The framework includes perception modules for transparent object segmentation and depth completion, which are key to the pipeline.
  edge-guided perception mechanismTransGraspNet-Depth completion network
• Deep Learningframework80%
  The perception and depth completion modules are based on deep learning architectures like Mask R-CNN and TDCNet.
  refine Mask R-CNN with CBAM attentionTDCNet backbone
• Computer Visionsubfield70%
  The paper addresses transparent object segmentation and depth reconstruction, which are computer vision tasks.
  transparent object segmentationdepth completion
• Convolutional Neural Networksframework70%
  The perception network uses convolutional neural networks (ResNet-101, FPN) for feature extraction.
  ResNet-101 + FPN backbone
• Robot Learningsubfield50%
  The system uses deep learning models for perception and depth completion, but the core contribution is the consistency framework rather than learning methodology.
  deep learning, data-driven methods became mainstreamfine-tuning on RobotSci-Glass Perception Subset

Abstract

Manipulating transparent laboratory glassware that contains liquid is inherently safety-critical: even small geometric errors can cause unstable grasps and hazardous spillage. Although recent progress has been made in transparent object perception and robotic grasping, most existing systems optimize detection, depth reconstruction, and grasp planning independently, which leads to cross-stage inconsistency imperfect boundaries induce depth bleeding, distorted surfaces corrupt normal estimation, and task agnostic grasp scoring yields tilted or off-center grasps that fail under dynamic motion. In this paper, we propose TransGraspNet, a geometry physics consistent framework that explicitly enforces consistency from perception to execution through three coupled principles: boundary consistency to produce structurally reliable object contours as downstream priors, surface consistency to preserve geometric fidelity and surface normal accuracy during depth reconstruction, and physics consistency to refine grasp selection with centroid alignment and wrench-space stability for upright and dynamically robust manipulation. We evaluate TransGraspNet on public benchmarks, a dedicated transparent glassware dataset, and a real robotic platform. The results show improved boundary quality and surface normal fidelity, and demonstrate strong task-level performance in cluttered transparent scenes. Most importantly, the proposed system achieves reliable real-world operation, including high grasp success rates in clutter and zero spillage during high speed liquid transport, highlighting the effectiveness of our method.

Paper Context