[{"data":1,"prerenderedAt":320},["ShallowReactive",2],{"project-detail-\u002Fprojects\u002Furgrasp-agent":3,"project-detail-nav":55},{"id":4,"title":5,"body":6,"category":22,"cover":23,"description":24,"extension":25,"highlights":26,"images":31,"links":32,"meta":33,"navigation":35,"path":36,"period":37,"portrait":38,"poster":31,"role":39,"seo":40,"stack":41,"status":47,"stem":48,"tags":49,"video":18,"videos":31,"__hash__":54},"projects\u002Fprojects\u002Furgrasp-agent.md","UR5 Grasp Agent：从手眼标定到 GraspNet 抓取",{"type":7,"value":8,"toc":17},"minimark",[9,13],[10,11,12],"h2",{"id":12},"项目简介",[14,15,16],"p",{},"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":18,"searchDepth":19,"depth":19,"links":20},"",2,[21],{"id":12,"depth":19,"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。","md",[27,28,29,30],"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。",null,{},{"date":34},"2025-09-01",true,"\u002Fprojects\u002Furgrasp-agent","2025.12 – 2026.04",false,"独立开发",{"title":5,"description":24},[42,43,44,45,46],"Python","PyTorch","OpenCV","GraspNet","ur_rtde","进行中","projects\u002Furgrasp-agent",[45,50,51,52,53],"UR5","手眼标定","RealSense","6D 抓取位姿","YyQkluZioC8fIDzM_lYdoYv_sstvJ1FHw1GVcy43mYU",[56,133,223,304],{"id":57,"title":58,"body":59,"category":103,"cover":104,"description":105,"extension":25,"highlights":106,"images":111,"links":117,"meta":119,"navigation":35,"path":121,"period":122,"portrait":38,"poster":31,"role":39,"seo":123,"stack":124,"status":47,"stem":127,"tags":128,"video":131,"videos":31,"__hash__":132},"projects\u002Fprojects\u002Far-ur5-real.md","真机拓展：Quest 3 AR 透视控制真实 UR5",{"type":7,"value":60,"toc":98},[61,63,66,69,92,95],[10,62,12],{"id":12},[14,64,65],{},"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。",[10,67,68],{"id":68},"技术方案",[70,71,72,80,86],"ul",{},[73,74,75,79],"li",{},[76,77,78],"strong",{},"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。",[73,81,82,85],{},[76,83,84],{},"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。",[73,87,88,91],{},[76,89,90],{},"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。",[10,93,94],{"id":94},"个人收获",[14,96,97],{},"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":18,"searchDepth":19,"depth":19,"links":99},[100,101,102],{"id":12,"depth":19,"text":12},{"id":68,"depth":19,"text":68},{"id":94,"depth":19,"text":94},"具身智能","\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。",[107,108,109,110],"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。",[112,114],{"src":104,"title":113},"真机 UR5、RobotiQ 夹爪与 D405 腕部相机",{"src":115,"title":116},"\u002Fimages\u002Fprojects\u002Freal-d405-mount.jpg","D405 相机法兰安装",{"github":118},"https:\u002F\u002Fgithub.com\u002Fwkcherish\u002FUR5_Teleop.git",{"date":120},"2026-09-01","\u002Fprojects\u002Far-ur5-real","2026.08 – 至今",{"title":58,"description":105},[125,42,50,126],"Quest 3","Three.js","projects\u002Far-ur5-real",[125,84,50,129,130],"UR5_Teleop","遥操作","\u002Fvideos\u002Fprojects\u002Far-ur5.mp4","Piwg_HpfiihFgsBjLUZdewuitikfE9K5-oasSLhNyGM",{"id":134,"title":135,"body":136,"category":103,"cover":187,"description":188,"extension":25,"highlights":189,"images":194,"links":209,"meta":210,"navigation":35,"path":212,"period":213,"portrait":38,"poster":31,"role":39,"seo":214,"stack":215,"status":218,"stem":219,"tags":220,"video":18,"videos":31,"__hash__":222},"projects\u002Fprojects\u002Far-ur5-rl.md","Quest 3 VR 远程操控 UR5 数字孪生（纯仿真）",{"type":7,"value":137,"toc":182},[138,140,143,145,177,179],[10,139,12],{"id":12},[14,141,142],{},"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。",[10,144,68],{"id":68},[70,146,147,153,159,165,171],{},[73,148,149,152],{},[76,150,151],{},"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。",[73,154,155,158],{},[76,156,157],{},"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。",[73,160,161,164],{},[76,162,163],{},"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。",[73,166,167,170],{},[76,168,169],{},"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。",[73,172,173,176],{},[76,174,175],{},"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。",[10,178,94],{"id":94},[14,180,181],{},"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":18,"searchDepth":19,"depth":19,"links":183},[184,185,186],{"id":12,"depth":19,"text":12},{"id":68,"depth":19,"text":68},{"id":94,"depth":19,"text":94},"\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。",[190,191,192,193],"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。",[195,197,200,203,206],{"src":187,"title":196},"MuJoCo 仿真场景：UR5、工作台与工具箱",{"src":198,"title":199},"\u002Fimages\u002Fprojects\u002Fsim-overview-2.png","仿真场景另一视角",{"src":201,"title":202},"\u002Fimages\u002Fprojects\u002Fsim-overview-3.png","工作台后侧视角",{"src":204,"title":205},"\u002Fimages\u002Fprojects\u002Fsim-d405-view.png","D405 腕部相机渲染视角",{"src":207,"title":208},"\u002Fimages\u002Fprojects\u002Fsim-cad-mount.png","相机法兰安装件设计",{},{"date":211},"2026-06-15","\u002Fprojects\u002Far-ur5-rl","2026.03 – 2026.06",{"title":135,"description":188},[216,126,217,125,42],"MuJoCo","FastAPI","已完成","projects\u002Far-ur5-rl",[216,126,217,125,221],"WebXR","kXCTEg8RB2d0GP8OHN0IfhCQg71WeWqr28NSnqb6jHU",{"id":224,"title":225,"body":226,"category":267,"cover":268,"description":269,"extension":25,"highlights":270,"images":31,"links":275,"meta":277,"navigation":35,"path":279,"period":280,"portrait":38,"poster":268,"role":281,"seo":282,"stack":283,"status":218,"stem":286,"tags":287,"video":291,"videos":292,"__hash__":303},"projects\u002Fprojects\u002Flerobot-vla-reproduction.md","LeRobot101：ACT \u002F smolVLA \u002F pi0.5 复现",{"type":7,"value":227,"toc":262},[228,230,233,236,256,259],[10,229,12],{"id":12},[14,231,232],{},"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。",[10,234,235],{"id":235},"复现内容",[70,237,238,244,250],{},[73,239,240,243],{},[76,241,242],{},"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。",[73,245,246,249],{},[76,247,248],{},"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。",[73,251,252,255],{},[76,253,254],{},"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。",[10,257,258],{"id":258},"评估结果",[14,260,261],{},"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":18,"searchDepth":19,"depth":19,"links":263},[264,265,266],{"id":12,"depth":19,"text":12},{"id":235,"depth":19,"text":235},{"id":258,"depth":19,"text":258},"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。",[271,272,273,274],"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":276},"https:\u002F\u002Fgithub.com\u002Fwkcherish\u002Flerobot.git",{"date":278},"2026-05-20","\u002Fprojects\u002Flerobot-vla-reproduction","2025.11 – 2026.05","独立复现",{"title":225,"description":269},[284,43,285,216],"LeRobot","Transformers","projects\u002Flerobot-vla-reproduction",[284,288,248,254,289,290],"ACT","模仿学习","VLA","\u002Fvideos\u002Fprojects\u002Fact.mp4",[293,295,299],{"src":291,"title":294,"poster":268},"ACT 策略仿真任务演示",{"src":296,"title":297,"poster":298},"\u002Fvideos\u002Fprojects\u002Fsmolvla.mp4","smolVLA 桌面任务复现","\u002Fimages\u002Fprojects\u002Fsmolvla-poster.jpg",{"src":300,"title":301,"poster":302},"\u002Fvideos\u002Fprojects\u002Fpi0.5.mp4","pi0.5 微调策略执行","\u002Fimages\u002Fprojects\u002Fpi05-poster.jpg","OBkLJ9rOpfpt66cMl1TkSBaef3E8ALjvOwncZv3t9WU",{"id":4,"title":5,"body":305,"category":22,"cover":23,"description":24,"extension":25,"highlights":314,"images":31,"links":315,"meta":316,"navigation":35,"path":36,"period":37,"portrait":38,"poster":31,"role":39,"seo":317,"stack":318,"status":47,"stem":48,"tags":319,"video":18,"videos":31,"__hash__":54},{"type":7,"value":306,"toc":311},[307,309],[10,308,12],{"id":12},[14,310,16],{},{"title":18,"searchDepth":19,"depth":19,"links":312},[313],{"id":12,"depth":19,"text":12},[27,28,29,30],{},{"date":34},{"title":5,"description":24},[42,43,44,45,46],[45,50,51,52,53],1791660993167]