# GRASP-006: A New Technique for Fully Autonomous and Efficient 3D Robotics Hand/Eye Calibration > 本页由本地论文卡片生成。请不要直接编辑本页；修改源卡片后运行 `vp run docs:generate`。 > 中文支持说明：本页是中文站点镜像，保留论文标题、证据状态和来源字段的原始表述，避免未经核实的翻译改变证据边界。 > 内部 wiki 边界：本页提供论文卡片证据和用户提供的 PDF 下载入口，仅用于私有阅读；全文抽取文本不发布。 > 来源: `library/paper_cards/GRASP-006.md` ## 论文访问 * 内部 PDF: 下载论文 PDF * DOI / 官方页面: [10.1109/70.34770](https://doi.org/10.1109/70.34770) * 部署边界: 这些 PDF 链接只适合私有/受保护的内部 wiki。 ## GRASP-006: A New Technique for Fully Autonomous and Efficient 3D Robotics Hand/Eye Calibration ## Metadata * ID: GRASP-006 * Title: A New Technique for Fully Autonomous and Efficient 3D Robotics Hand/Eye Calibration * Year: 1989 * DOI / URL: 10.1109/70.34770 * Local PDF: 见上方论文访问区块 * Text artifact: local-only path withheld from docs site * Review status: `extracted from OCR` ## Study Type * Track: Robot Vision / Grasping / Calibration * Task: fully autonomous 3D robot hand/eye calibration for an eye-on-hand camera * Participants or dataset: simulation experiments and real robot experiments; no human participants * Hardware: IBM Clean Room Robot, rigidly mounted Javelin CCDE 480 x 388 CCD camera, calibration block with 36 printed discs * Channels or sensors: monocular camera images, robot joint/pose readings, calibration target feature coordinates ## Methods * Paradigm: estimate the 3D homogeneous transform between camera coordinate frame and robot gripper coordinate frame * Signal processing or model: decoupled hand/eye calibration using interstation robot and camera motions; solve rotation and translation with linear least-squares systems over station pairs * Training/calibration: robot automatically moves a camera rigidly mounted to the gripper through multiple stations, captures a calibration block image at each pause, estimates camera extrinsics per station, then solves camera-to-gripper transform * Online/offline: calibration procedure validated through simulation and real robot experiments; not an online grasping controller ## Results * Metrics: computation operations, image/extrinsic calibration time per station, new camera pose prediction rotation and translation errors * Main findings: per-station image acquisition, feature extraction, and camera extrinsic calibration takes about 90 ms; hand/eye computation after robot motion takes roughly 100 + 60N arithmetic operations; real experiments report lower pose prediction error as station count increases, with about 2.888 mrad rotation error and 14.642 mil translation error for 10 stations * Reported limitations: accuracy depends strongly on interstation rotation angle, angle between interstation rotation axes, camera-to-calibration-block distance, gripper-center distance across stations, robot orientation/translation errors, and calibration feature extraction accuracy; OCR text contains some recognition errors in equations and symbols ## Relevance To This Project * Supports: the need for a camera-to-gripper transform when converting camera-frame visual measurements into robot-frame actions * Conflicts with: does not evaluate modern RGB-D grasping, YOLO, BCI, shared autonomy, or the actual `SAH-BRI-Grasp` hardware * Design implication: `SAH-BRI-Grasp` can cite this paper for classical eye-on-hand calibration rationale and station-planning factors, but still must document and validate its own camera/robot calibration procedure ## Extracted Evidence | Claim | Status | Evidence Note | Page/Section | | --- | --- | --- | --- | | Hand/eye calibration estimates the relative 3D pose between a camera and the robot gripper in an eye-on-hand setup. | verified | The abstract and Section I.B define the task as computing camera position/orientation relative to the last joint or gripper. | Abstract; Section I.B | | Camera-to-gripper calibration is needed to relate vision measurements to robot motion for grasping or assembly. | verified | The introduction says that without hand/eye calibration, a vision system may know object pose relative to the sensor but the robot does not know how to place the manipulator to grasp it. | Section I.D | | The proposed method uses multiple robot/camera stations and solves rotation and translation from interstation motion relationships. | verified | The method defines gripper/camera frames, station transforms, the minimum station requirements, and linear systems for rotation and translation. | Section II | | Accuracy improves when interstation rotation axes and angles are well chosen and redundant stations are used. | verified | The accuracy section lists critical factors and recommends maximizing rotation-axis separation, maximizing interstation rotation, and using redundant stations. | Section III.B-III.C | | OCR recovery succeeded, but equations and symbols require manual caution. | verified | The local text artifact was recovered through OCR on 2026-07-10; body text is readable, while some mathematical notation is noisy. | Text artifact | | Project-specific hand-eye calibration remains unverified. | needs confirmation | The paper supports calibration rationale and design factors, but it does not document the actual SAH-BRI-Grasp camera, robot, target, procedure, or validation. | Relevance assessment | ## Open Questions * Which camera, calibration target, station count, and robot motion plan will the SAH-BRI-Grasp prototype use? * Should the prototype use a classical Tsai-Lenz-style hand-eye calibration routine, an OpenCV hand-eye solver, or a learned/servoing alternative? * What acceptance metric should be used before robot grasp execution: reprojection error, held-out pose prediction, grasp-frame consistency, or task-level grasp success?