Robotic grasping of hollow guardrail components for outdoor installation: a geometry-based RGB-D depth sampling approach for 3D localization

Abstract

Purpose This study aims to develop an integrated sensing–perception–control system for automated highway guardrail installation, addressing the depth perception failure of RGB-D cameras over hollow structural regions in field working conditions. Design/methodology/approach An edge-exterior multilayer depth sampling strategy with 72 points on three concentric rings is proposed to obtain reliable surface depth from the solid body surrounding the hollow aperture under oblique viewing and partial occlusion. A lightweight segmentation model, YOLOv11n-PMS-seg, adopts a feature optimization–attention enhancement–shape constraint design, supported by ellipse fitting, Kalman filtering, track-ID recovery and adaptive CAN transmission. A one-time teaching step records the relative pose between the gripper and hole; the manipulator servos to this reference, eliminating explicit hand-eye calibration and making grasping invariant to forklift position and guardrail height. Findings The depth acquisition success rate reaches 99.4%. Under a single calibration at 1,250 mm, localization error is 17.8 ± 4.2 mm at 30 fps. Across 200 outdoor winter trials (−5 to 5 °C), 194 were successful; the six failures manifested as system-level freezes rather than as erroneous perception output. For low-illumination indoor scenes, a 6 W LED supplementary light maintains depth quality. YOLOv11n-PMS-seg cuts parameters by 7.8% and GFLOPs by 10.7% while improving mAP@0.5:0.95 by 3.3 percentage points. Research limitations/implications The system is currently validated at a fixed vertical distance of 1,250 mm for single-target scenarios. Failure attribution relied on field observation, as automated telemetry logging was not implemented. Future work should address variable operating distances, multi-target grasping and onboard failure diagnosis to broaden deployment applicability. Practical implications The proposed system enables automated robotic grasping of hollow guardrail components with a 97.0% success rate under real outdoor conditions, directly reducing manual labor and safety risks in highway construction. Its lightweight architecture (17% GPU utilization) allows integration into existing industrial vehicle platforms without significant hardware upgrades, offering a cost-effective solution for infrastructure automation. Social implications Automating highway guardrail installation reduces worker exposure to traffic hazards and physically demanding labor in road construction environments. Validated outdoor winter deployment demonstrates the feasibility of year-round infrastructure maintenance, contributing to safer roads and more efficient use of construction resources in civil engineering applications. Originality/value This work presents an integrated grasping pipeline for hollow guardrail components that combines a geometry-based depth sampling strategy with a teaching-based relative-pose scheme, without relying on additional learning-based depth completion networks. Validation covers winter outdoor daytime conditions at a single calibration distance; rainfall, direct sunlight and other guardrail specifications remain future work.