A High-Precision Perception Framework for Robotic Grasping Based on 3D Point Cloud Pose Estimation

Accurate 6D pose estimation using consumer-grade depth cameras in real-world scenarios

WaveletkAN Pose Refinement for Robot Bin Picking Driven by Cluttered RGB-D Point Clouds

Grasping cluttered boxes is still limited by the gap between rough 6-degree-of-freedom perception and the millimeter-level alignment required for robot execution. WaveletkAN is a pose-refinement model based on RGB-D point clouds proposed in this paper to correct object pose hypotheses for dense bins with occlusion, specular depth noise, and self-similar industrial parts. The method represents each candidate as a local observed point set, a CAD-derived canonical set, and a compact bin-context tensor, then predicts a residual rigid motion via Kolmogorov-Arnold network layers parametrized by learnable wavelet atoms. A confidence-weighted correspondence field and a residual SE (3) update can be used with the traditional proposal generator and grasp planner. WaveletkAN reduced the median translation error from 7.8 mm to 2.9 mm and the median rotation error from 5.6 degrees to 1.9 degrees after refinement in a simulated-real mixed benchmark with 18 object categories and 42,600 evaluated hypotheses. The successful pick rate in dense clutter rose to 93.1 %, and the mean refinement latency of the industrial GPU was still 11.6 ms. Ablation experiments showed that removing the wavelet basis increased ADD-S by 31.7%, and omitting context gating reduced top-1 executable pose recall by 6.8 %. Based on the above experiments, multi-scale functional parameterization can achieve robust and deployable pose correction for RGB-D robotic bin-picking systems.

Two-stage Monocular 6D Pose Estimation for Small Cubic Objects

This paper studies monocular 6D pose estimation of small cubic objects from a single RGB image and proposes a two-stage manipulation- oriented framework, which achieves the strongest overall balance in ADD-S, translation accuracy, rotation stability, and task-oriented usability metrics.

An integrated 3D vision sensor system with adaptive point cloud processing for robotic workpiece localization in industrial bin picking

The transition from automated to intelligent manufacturing increasingly relies on three-dimensional (3D) machine vision for robot guidance. Nevertheless, existing 3D vision systems still suffer from long processing delays, poor adaptability to changing environments, and inadequate pose accuracy in real industrial settings. To address these issues, this paper proposes a real-time point cloud processing and workpiece localization system that integrates multi-module optimization with a dynamic adaptive framework. The system acquires data through binocular stereo vision and incorporates optimized spatial filtering, PCA-based dimensionality reduction, and a machine learning-enhanced FAST feature detector. A hand-eye calibration model that includes distortion compensation achieves sub-millimeter mapping from image coordinates to the robot workspace. Experimental results on the public LineMOD dataset and a custom industrial bin-picking dataset show that the proposed system attains a 99% grasping success rate under favorable lighting and 96% under challenging conditions, with an average cycle time of 520 ms. Translation error reaches 0.38 mm under good lighting (rotation error: 0.61°, ADD: 0.47 mm).Comparative evaluations confirm substantial gains in speed, accuracy, and environmental robustness relative to existing commercial and academic systems. Furthermore, the system sustains reliable performance under Gaussian noise up to 0.6 mm and occlusion levels up to 40%, confirming its viability for high-throughput industrial applications.

A Low-Cost 3D Rigid Object Posture Estimation Using 2D LiDAR Dynamics Elevation Mapping System for Autonomous Vehicle

This research presents the development of a low-cost 3D rigid-object posture estimation system using a 2D LiDAR sensor with dynamic elevation mapping for autonomous vehicle applications. The methodology encompasses four key components: (1) the development of a data acquisition system that integrates a 2D LiDAR with a servo-controlled elevation mechanism, enabling precise vertical scanning; (2) the implementation of coordinate transformation algorithms to reconstruct accurate 3D point clouds from sequential 2D scans; (3) the optimization of point cloud density through multi-parameter approaches, including adaptive scan resolution and noise filtering; and (4) the application of the Iterative Closest Point (ICP) algorithm for precise object posture estimation, ensuring robust alignment between observed and reference point clouds. The system is designed to enhance perception in autonomous driving by providing real-time, high-accuracy 3D pose estimation while maintaining affordability and computational efficiency. The system's performance was evaluated through testing in two distinct environments: an indoor laboratory setting (Room PS 03.07) and a corridor space (SAW Building) at PENS campus Sukolilo, demonstrating the system's capability to generate accurate 3D representations with color-coded elevation mapping ranging from 0.00 to 8.48 meters, while the point cloud optimization achieved efficient data compression through a voxel grid filter with 4mm leaf size, ensuring optimal point density between 4-8mm minimum point distances, successfully detecting and mapping various rigid objects while maintaining geometric accuracy in both confined and extended spaces. The results demonstrate cost-effective implementations by employing alternative sensing systems rather than relying on 3D LiDAR.

M-VTOP: Modular Visuo-Tactile Object Pose Estimation for High-Precision Robotic Manipulation

Experiments show that M-VTOP achieves sub-millimeter accuracy under complex geometries, occlusions, and tight tolerances, demonstrating its promise for high-precision robotic manipulation.