[{"data":1,"prerenderedAt":356},["ShallowReactive",2],{"project-detail-\u002Fprojects\u002Far-ur5-rl":3,"project-detail-nav":113},{"id":4,"title":5,"body":6,"category":68,"cover":69,"description":70,"extension":71,"highlights":72,"images":77,"links":92,"meta":93,"navigation":95,"path":96,"period":97,"portrait":98,"poster":99,"role":100,"seo":101,"stack":102,"status":108,"stem":109,"tags":110,"video":62,"videos":99,"__hash__":112},"projects\u002Fprojects\u002Far-ur5-rl.md","Quest 3 VR 远程操控 UR5 数字孪生（纯仿真）",{"type":7,"value":8,"toc":61},"minimark",[9,13,17,20,55,58],[10,11,12],"h2",{"id":12},"项目简介",[14,15,16],"p",{},"搭建可用于交互控制和状态反馈的 UR5 机械臂 MuJoCo 数字孪生仿真。前端基于 Three.js 构建与后端状态相互映射的 3D 可视化界面，后端使用 FastAPI 接收控制输入并驱动仿真；Quest 3 通过局域网 WebXR 进入 VR 后，可用手柄或手势连续控制机械臂与夹爪。本项目为纯仿真工作，真机 AR 拓展为独立项目。",[10,18,19],{"id":19},"技术方案",[21,22,23,31,37,43,49],"ul",{},[24,25,26,30],"li",{},[27,28,29],"strong",{},"仿真环境","：MuJoCo 中建立 UR5 机械臂、工作台、工具箱与工具模型，配置接触、摩擦与执行器参数。",[24,32,33,36],{},[27,34,35],{},"可视化","：Three.js 实时渲染机械臂状态，与后端通过 WebSocket\u002FHTTP 保持状态同步。",[24,38,39,42],{},[27,40,41],{},"通信层","：FastAPI 提供状态观测、相机画面与控制接口，桥接前端输入与仿真动作。",[24,44,45,48],{},[27,46,47],{},"VR 交互","：Quest 3 利用手柄 6DoF 位姿或 Hand Tracking 手势，经笛卡尔伺服映射为末端执行器目标位姿。",[24,50,51,54],{},[27,52,53],{},"伺服控制","：后端通过 MuJoCo Jacobian 将末端误差换算为关节速度，并串联限位、奇异点降速与碰撞保护。",[10,56,57],{"id":57},"个人收获",[14,59,60],{},"系统理解了数字孪生系统从建模、通信到 VR 交互的完整链路，掌握了 WebXR 开发与基于 Jacobian 的笛卡尔伺服方法，也为后续真机遥操作与模仿学习复现打下基础。",{"title":62,"searchDepth":63,"depth":63,"links":64},"",2,[65,66,67],{"id":12,"depth":63,"text":12},{"id":19,"depth":63,"text":19},{"id":57,"depth":63,"text":57},"具身智能","\u002Fimages\u002Fprojects\u002Fsim-overview.png","基于 MuJoCo 的 UR5 数字孪生仿真，Three.js 可视化与 FastAPI 后端，Quest 3 通过 WebXR 在 VR 中以手柄或手势控制机械臂。","md",[73,74,75,76],"在 MuJoCo 中搭建 UR5 机械臂、工作台与工具箱、钳子、扳手等工具的高保真仿真场景，FastAPI 后端提供状态观测与控制接口。","Three.js 构建与后端状态实时双向映射的 3D 数字孪生界面，桌面浏览器即可完成调试与观测。","Quest 3 经局域网以 WebXR 进入 VR，手柄或手势通过 6DoF 笛卡尔伺服控制机械臂，Trigger 连续控制夹爪开合。","后端基于 Jacobian 实现 Cartesian servo，并施加速度、加速度、关节限位与奇异点约束，保证仿真中运动平滑安全。",[78,80,83,86,89],{"src":69,"title":79},"MuJoCo 仿真场景：UR5、工作台与工具箱",{"src":81,"title":82},"\u002Fimages\u002Fprojects\u002Fsim-overview-2.png","仿真场景另一视角",{"src":84,"title":85},"\u002Fimages\u002Fprojects\u002Fsim-overview-3.png","工作台后侧视角",{"src":87,"title":88},"\u002Fimages\u002Fprojects\u002Fsim-d405-view.png","D405 腕部相机渲染视角",{"src":90,"title":91},"\u002Fimages\u002Fprojects\u002Fsim-cad-mount.png","相机法兰安装件设计",{},{"date":94},"2026-06-15",true,"\u002Fprojects\u002Far-ur5-rl","2026.03 – 2026.06",false,null,"独立开发",{"title":5,"description":70},[103,104,105,106,107],"MuJoCo","Three.js","FastAPI","Quest 3","Python","已完成","projects\u002Far-ur5-rl",[103,104,105,106,111],"WebXR","kXCTEg8RB2d0GP8OHN0IfhCQg71WeWqr28NSnqb6jHU",[114,185,237,319],{"id":115,"title":116,"body":117,"category":68,"cover":156,"description":157,"extension":71,"highlights":158,"images":163,"links":169,"meta":171,"navigation":95,"path":173,"period":174,"portrait":98,"poster":99,"role":100,"seo":175,"stack":176,"status":178,"stem":179,"tags":180,"video":183,"videos":99,"__hash__":184},"projects\u002Fprojects\u002Far-ur5-real.md","真机拓展：Quest 3 AR 透视控制真实 UR5",{"type":7,"value":118,"toc":151},[119,121,124,126,146,148],[10,120,12],{"id":12},[14,122,123],{},"在「Quest 3 VR 远程操控 UR5 数字孪生」纯仿真项目的基础上，独立开发真机控制链路 UR5_Teleop：Quest 3 以 AR 彩色透视模式运行，将头显与手柄位姿映射到真实 UR5 机械臂，在实际工作台上完成遥控运动。该拓展与仿真仓库代码相互独立。",[10,125,19],{"id":19},[21,127,128,134,140],{},[24,129,130,133],{},[27,131,132],{},"真机通信","：头显控制位姿经局域网发送至真机控制服务，转换为 UR5 可执行的末端目标。",[24,135,136,139],{},[27,137,138],{},"AR 透视","：利用 Quest 3 彩色透视直接观察真实工作台与机械臂，无需重建场景模型。",[24,141,142,145],{},[27,143,144],{},"控制迁移","：复用仿真中验证的笛卡尔伺服与夹爪连续控制逻辑，缩短真机调试周期。",[10,147,57],{"id":57},[14,149,150],{},"完整走通了控制策略从仿真到真机的迁移过程，体会到 AR 透视在遥操作中降低空间认知负担的作用，也积累了真机安全控制与通信时延处理的经验。",{"title":62,"searchDepth":63,"depth":63,"links":152},[153,154,155],{"id":12,"depth":63,"text":12},{"id":19,"depth":63,"text":19},{"id":57,"depth":63,"text":57},"\u002Fimages\u002Fprojects\u002Freal-hardware-overview.jpg","在纯仿真项目基础上独立拓展真机控制链路 UR5_Teleop，Quest 3 AR 彩色透视下将头显与手柄位姿映射到真实 UR5 完成遥控运动。",[159,160,161,162],"在纯仿真项目基础上独立拓展真机控制链路（UR5_Teleop），将 Quest 3 头显与手柄位姿映射到真实 UR5。","Quest 3 AR 彩色透视下保留真实工作台视野，在实际工作台上完成机械臂遥控运动。","真机端复用笛卡尔伺服与连续夹爪控制逻辑，仿真中验证的控制链路可直接向真机迁移。","拓展代码 UR5_Teleop 与仿真仓库相互独立，已开源。",[164,166],{"src":156,"title":165},"真机 UR5、RobotiQ 夹爪与 D405 腕部相机",{"src":167,"title":168},"\u002Fimages\u002Fprojects\u002Freal-d405-mount.jpg","D405 相机法兰安装",{"github":170},"https:\u002F\u002Fgithub.com\u002Fwkcherish\u002FUR5_Teleop.git",{"date":172},"2026-09-01","\u002Fprojects\u002Far-ur5-real","2026.08 – 至今",{"title":116,"description":157},[106,107,177,104],"UR5","进行中","projects\u002Far-ur5-real",[106,138,177,181,182],"UR5_Teleop","遥操作","\u002Fvideos\u002Fprojects\u002Far-ur5.mp4","Piwg_HpfiihFgsBjLUZdewuitikfE9K5-oasSLhNyGM",{"id":4,"title":5,"body":186,"category":68,"cover":69,"description":70,"extension":71,"highlights":225,"images":226,"links":232,"meta":233,"navigation":95,"path":96,"period":97,"portrait":98,"poster":99,"role":100,"seo":234,"stack":235,"status":108,"stem":109,"tags":236,"video":62,"videos":99,"__hash__":112},{"type":7,"value":187,"toc":220},[188,190,192,194,216,218],[10,189,12],{"id":12},[14,191,16],{},[10,193,19],{"id":19},[21,195,196,200,204,208,212],{},[24,197,198,30],{},[27,199,29],{},[24,201,202,36],{},[27,203,35],{},[24,205,206,42],{},[27,207,41],{},[24,209,210,48],{},[27,211,47],{},[24,213,214,54],{},[27,215,53],{},[10,217,57],{"id":57},[14,219,60],{},{"title":62,"searchDepth":63,"depth":63,"links":221},[222,223,224],{"id":12,"depth":63,"text":12},{"id":19,"depth":63,"text":19},{"id":57,"depth":63,"text":57},[73,74,75,76],[227,228,229,230,231],{"src":69,"title":79},{"src":81,"title":82},{"src":84,"title":85},{"src":87,"title":88},{"src":90,"title":91},{},{"date":94},{"title":5,"description":70},[103,104,105,106,107],[103,104,105,106,111],{"id":238,"title":239,"body":240,"category":281,"cover":282,"description":283,"extension":71,"highlights":284,"images":99,"links":289,"meta":291,"navigation":95,"path":293,"period":294,"portrait":98,"poster":282,"role":295,"seo":296,"stack":297,"status":108,"stem":301,"tags":302,"video":306,"videos":307,"__hash__":318},"projects\u002Fprojects\u002Flerobot-vla-reproduction.md","LeRobot101：ACT \u002F smolVLA \u002F pi0.5 复现",{"type":7,"value":241,"toc":276},[242,244,247,250,270,273],[10,243,12],{"id":12},[14,245,246],{},"以 Hugging Face LeRobot 框架为主线，系统复现 ACT、smolVLA、pi0.5 三个代表性模型，覆盖模仿学习与视觉-语言-动作两大技术路线。项目打通了从遥操作数据采集、策略训练到仿真评估的完整链路，所有复现代码已整理上传至 GitHub。",[10,248,249],{"id":249},"复现内容",[21,251,252,258,264],{},[24,253,254,257],{},[27,255,256],{},"ACT（Action Chunking with Transformers）","：以动作分块预测替代单步动作回归，配合时间集成（temporal ensembling）平滑执行轨迹，复现数据采集、训练与仿真评估流程。",[24,259,260,263],{},[27,261,262],{},"smolVLA","：基于 SmolVLM 视觉-语言骨干的轻量 VLA 模型，在桌面操作任务上完成端到端复现，理解视觉观测与语言指令到动作 token 的映射方式。",[24,265,266,269],{},[27,267,268],{},"pi0.5","：基于 OpenPI 的预训练 VLA 策略，在自有任务配置上进行微调，验证「语言指令 → 视觉-语言理解 → 动作序列输出」的完整执行链路。",[10,271,272],{"id":272},"评估结果",[14,274,275],{},"在 Robosuite 中以脚本专家生成演示数据，训练行为克隆（BC）策略；在连续 20 轮评估中达到 80% 任务成功率，验证了数据质量与动作分块对策略稳定性的作用。",{"title":62,"searchDepth":63,"depth":63,"links":277},[278,279,280],{"id":12,"depth":63,"text":12},{"id":249,"depth":63,"text":249},{"id":272,"depth":63,"text":272},"机器人学习","\u002Fimages\u002Fprojects\u002Fact-poster.jpg","基于 LeRobot 框架复现 ACT、smolVLA 与 pi0.5 三类模仿学习与视觉-语言-动作模型，打通数据采集、策略训练到仿真评估全流程。",[285,286,287,288],"跟随 LeRobot101 复现 ACT 模型，理解动作分块（Action Chunking）与模仿学习训练流程。","复现 smolVLA 轻量 VLA 模型，在仿真桌面任务上完成数据采集、策略训练与评估。","复现 pi0.5：基于 OpenPI 预训练 VLA 策略进行微调，验证语言指令到机械臂动作的执行链路。","配合 Robosuite 行为克隆实验，从脚本专家数据训练策略，20 轮评估达到 80% 成功率。",{"github":290},"https:\u002F\u002Fgithub.com\u002Fwkcherish\u002Flerobot.git",{"date":292},"2026-05-20","\u002Fprojects\u002Flerobot-vla-reproduction","2025.11 – 2026.05","独立复现",{"title":239,"description":283},[298,299,300,103],"LeRobot","PyTorch","Transformers","projects\u002Flerobot-vla-reproduction",[298,303,262,268,304,305],"ACT","模仿学习","VLA","\u002Fvideos\u002Fprojects\u002Fact.mp4",[308,310,314],{"src":306,"title":309,"poster":282},"ACT 策略仿真任务演示",{"src":311,"title":312,"poster":313},"\u002Fvideos\u002Fprojects\u002Fsmolvla.mp4","smolVLA 桌面任务复现","\u002Fimages\u002Fprojects\u002Fsmolvla-poster.jpg",{"src":315,"title":316,"poster":317},"\u002Fvideos\u002Fprojects\u002Fpi0.5.mp4","pi0.5 微调策略执行","\u002Fimages\u002Fprojects\u002Fpi05-poster.jpg","OBkLJ9rOpfpt66cMl1TkSBaef3E8ALjvOwncZv3t9WU",{"id":320,"title":321,"body":322,"category":332,"cover":333,"description":334,"extension":71,"highlights":335,"images":99,"links":340,"meta":341,"navigation":95,"path":343,"period":344,"portrait":98,"poster":99,"role":100,"seo":345,"stack":346,"status":178,"stem":350,"tags":351,"video":62,"videos":99,"__hash__":355},"projects\u002Fprojects\u002Furgrasp-agent.md","UR5 Grasp Agent：从手眼标定到 GraspNet 抓取",{"type":7,"value":323,"toc":329},[324,326],[10,325,12],{"id":12},[14,327,328],{},"构建一套从硬件标定到抓取策略的 UR5 抓取系统，打通 RealSense 感知、GraspNet 位姿检测与机械臂执行链路。",{"title":62,"searchDepth":63,"depth":63,"links":330},[331],{"id":12,"depth":63,"text":12},"机器人感知","\u002Fimages\u002Fprojects\u002Furgrasp-annotation.png","面向真实 UR5 的抓取智能体，覆盖相机标定、手眼标定、GraspNet 抓取位姿检测与机械臂运动执行全链路。",[336,337,338,339],"完成 RealSense 相机内参标定与眼在手上（eye-in-hand）手眼标定，统一米制与旋转矢量表示。","基于 GraspNet 从 RGB-D 场景检测 6D 抓取位姿，融合桌面物体分割与抓取点生成。","通过 ur_rtde 驱动 UR5 执行抓取运动，构建从视觉感知到运动执行的完整抓取 Agent。","统一相机系、标定板系与机器人基坐标系的米制尺度与旋转矢量表示，控制手眼标定重投影与位姿传递误差。",{},{"date":342},"2025-09-01","\u002Fprojects\u002Furgrasp-agent","2025.12 – 2026.04",{"title":321,"description":334},[107,299,347,348,349],"OpenCV","GraspNet","ur_rtde","projects\u002Furgrasp-agent",[348,177,352,353,354],"手眼标定","RealSense","6D 抓取位姿","YyQkluZioC8fIDzM_lYdoYv_sstvJ1FHw1GVcy43mYU",1791660993166]