KnowledgeGraph.canvas

{
	"nodes":[
		{"id":"title_card","type":"text","x":-500,"y":-3600,"width":1000,"height":160,"color":"6","text":"# 🧠 灵巧操作知识图谱 — Dexterous Manipulation Knowledge Graph\n\n**研究者**: PKU 灵巧操作方向 | **核心项目**: DNPM\n**结构**: 🔬Exp → 💡Ideas → 🚀Projects → ⚡Breakthroughs → 📄Papers → 🧠Foundations"},
		{"id":"exp_status_group","type":"group","x":-1600,"y":-3300,"width":500,"height":520,"color":"5","label":"🔬 实验验证状态"},
		{"id":"exp2_findings","type":"text","x":-1560,"y":-3250,"width":420,"height":430,"color":"5","text":"## 📊 Exp2 核心发现 (2026-02)\n\n### TA (Thumbaround)\n✅ **Light BASE SR=0.83** 最优\n❌ Heavy SR=0 (reward hacking)\n❌ TWC 无显著优势\n\n### TP (Triangle Pass)\n✅ **Medium TWC SR=0.86** 最优\n✅ TWC 降方差 19×\n❌ Heavy SR=0\n\n### ❗ 核心洞见\n**TA与TP对TWC/奖励响应完全不对称**\n→ 需任务特化策略"},
		{"id":"insights_group","type":"group","x":-1600,"y":-2500,"width":3220,"height":500,"color":"3","label":"💡 Research Insights: 原创 Idea (痛点-理论-文献三角定位)"},
		{"id":"idea_001","type":"text","x":-1560,"y":-2440,"width":340,"height":240,"color":"3","text":"## Idea-001 [P0]\n**Phase-Adaptive Impedance**\n\n💡 逐指时变阻抗 + 频率自适应\nsnap→高刚度 / spin→低刚度\n\n📊 历史Kp最优3.5~8.5 (TP)\n🎯 基线: TP Medium TWC SR=0.86"},
		{"id":"idea_002","type":"text","x":-1170,"y":-2440,"width":340,"height":220,"color":"3","text":"## Idea-002 [P0]\n**Autoregressive Exploration**\n\n💡 接触自适应 AR-p 探索噪声\n替代白噪声 → 时间相关探索\n\n⭐ Stage 0: β Grid Search"},
		{"id":"idea_007","type":"text","x":-780,"y":-2440,"width":340,"height":260,"color":"3","text":"## Idea-007 [P0]\n**Dual Orthogonal Curriculum**\n\n💡 α物理 × δ状态 双正交课程\n\n⚡ TWC不对称: TP有效/TA无效!\n📊 TP Reduced TWC方差↓19×\n🎯 TA需状态轴课程而非TWC"},
		{"id":"idea_006","type":"text","x":-390,"y":-2440,"width":340,"height":260,"color":"3","text":"## Idea-006 [P0→P1]\n**Adaptive Lipschitz Actor**\n\n💡 状态自适应 Lipschitz 约束\n稳相平滑 / 动态相锐利\n\n⚠️ Heavy SR=0 → 疑似动作抖动\n🎯 ALA验证: Heavy上能否恢复?"},
		{"id":"idea_003","type":"text","x":0,"y":-2440,"width":340,"height":260,"color":"3","text":"## Idea-003 [P1]\n**Causal Mediator Reward**\n\n💡 物理中介变量 → 因果奖励\n\n⚠️ Heavy SR=0 (reward hacking!)\n✅ TA Light BASE SR=0.83\n🎯 下一步: 单mediator(ω)测试"},
		{"id":"idea_004","type":"text","x":390,"y":-2440,"width":340,"height":220,"color":"3","text":"## Idea-004 [P1]\n**Convex Safe Set**\n\n💡 成功经验凸包\n几何 Bootstrapping\n\n⭐ 高惯性状态引导"},
		{"id":"idea_005","type":"text","x":780,"y":-2440,"width":340,"height":220,"color":"3","text":"## Idea-005 [P2]\n**Test-Time Contact Adapt**\n\n💡 在线接触参数辨识\n诊断性交互 + 策略适应\n\n⭐ 部署时在线校正"},
		{"id":"idea_combo","type":"text","x":1170,"y":-2440,"width":340,"height":260,"color":"1","text":"## ⚡ 最强组合\n\n**DOC + CA-ARP + ALA**\n双轴课程 + 相关噪声 + 平滑策略\n\n📊 Exp2核心: TA/TP不对称!\nTA→简洁奖励+BASE\nTP→中等奖励+TWC\n\n→ 见 [[_InsightsIndex]]"},
		{"id":"proj_dnpm_core","type":"group","x":-700,"y":-1700,"width":1900,"height":700,"color":"1","label":"🚀 Projects: Dynamic Non-Prehensile Manipulation"},
		{"id":"proj_dnpm","type":"file","file":"Projects/Dynamic Non-Prehensile Manipulation/Dynamic Non-Prehensile Manipulation.md","subpath":"#1. 摘要","x":-650,"y":-1640,"width":520,"height":280},
		{"id":"proj_dnpm_insight","type":"text","x":0,"y":-1640,"width":620,"height":280,"color":"1","text":"## 🎯 DNPM 核心洞见\n\n**惯性因果链条**:\n主动发力 → 高惯性状态 → 惯性力 → 对抗重力 → 物理演化\n\n**核心挑战**: 欠驱动系统的长程探索\n\n**任务**: Thumbaround (TA) · Triangle Pass (TP)\n**硬件**: UR5 + 灵巧手 | **仿真**: Isaac Gym\n**方法**: HDC (α-scaling 速度缩放)"},
		{"id":"proj_roadmap","type":"file","file":"Projects/Dynamic Non-Prehensile Manipulation/ideas.md","subpath":"#一、Big Picture：动态非紧握操作的任务定义、意义与本质困难","x":-650,"y":-1300,"width":520,"height":220},
		{"id":"proj_exp_status","type":"text","x":0,"y":-1300,"width":620,"height":220,"color":"5","text":"## 📊 实验进度\n\n| 实验 | 状态 | 关键发现 |\n|------|------|----------|\n| Smoke Test | ✅ | 8卡并行正常 |\n| Exp2: TA 奖励搜索 | ✅ | Light BASE SR=0.83 |\n| Exp2: TP 奖励搜索 | ✅ | Medium TWC SR=0.86 |\n| Exp3a: Alpha 训练 | 🔄 | 运行中 |"},
		{"id":"proj_wmts_group","type":"group","x":1300,"y":-1700,"width":1380,"height":1080,"color":"1","label":"🚀 Projects: World Model as Task Scheduler"},
		{"id":"proj_wmts_core","type":"file","file":"Projects/World Model as Task Scheduler/Final_WMTS.md","subpath":"#四、 动力学世界模型 (Ensemble World Model)","x":1350,"y":-1640,"width":400,"height":280},
		{"id":"proj_wmts_papers","type":"file","file":"Projects/World Model as Task Scheduler/RelatedPapersRecap/_RelatedPapersIndex.md","x":1800,"y":-1640,"width":350,"height":280},
		{"id":"proj_wmts_reliability","type":"file","file":"Projects/World Model as Task Scheduler/WMTS_Reliability_Extensions.md","x":2200,"y":-1640,"width":420,"height":280},
		{"id":"proj_wmts_insight","type":"text","x":1350,"y":-1300,"width":800,"height":370,"color":"1","text":"## 🔧 WMTS 五模块流水线\n\n**§一** 隐空间任务生成 (VAE+CMA-ES)\n**§二** Oracle 专才 (PPO+特权)\n**§三** Generalist 通才 (Diffusion Policy+CFG)\n**§四** Ensemble WM (Act Model + Rigid Model)\n**§五** 真机闭环微调 (Safety Checker + AWAC)\n\n**核心文件**: [[FOC_Control]] [[Actuator2RigidDynamicsModel_gap]]\n**RelatedPapers**: 40篇 (7类: WM核心/灵巧操作/Sim2Real/扩散/课程/理论/综述)\n\n**关键创新**: 串行因果链+梯度双通道 | L25 SDK→CAN→MCU 时序 | 温度级联漂移 | 转矩-速度椭圆包络"},
		{"id":"proj_wmts_insights_index","type":"file","file":"Projects/World Model as Task Scheduler/all_Insights_local/_InsightsIndex.md","x":2200,"y":-1300,"width":420,"height":370,"color":"4"},
		{"id":"proj_wmts_insights_summary","type":"text","x":1350,"y":-900,"width":1270,"height":280,"color":"4","text":"## 🔬 真机 RL Idea 队列 (15 ideas, 2026-04-27)\n\n**P0 立即启动 (5)**: TAR 无 GT pose reward · Latency-Aware Actuator · Failure-Mode Curriculum · WM-Guided Diffusion · PA-PER\n**P1 下一篇候选**: SBAL 主动数据收集 · ICHA 零梯度适应 · IECW 接触建模 · EBM 漂移检测 · WMID off-policy · WPTE 触觉编码 · SSMS · WG-ADR\n**P2 中期**: VQ Discrete Task Tokens · Reset-Free Autonomy\n\n**三大主线**: 真机 reward/data efficiency · Sim-to-Real 物理诊断 · 真机自主+test-time 适应\n\n**推荐论文组合**: Paper A (TAR+Reset-Free+WPTE → 完整真机系统) · Paper B (LAAA+PA-PER+WG-ADR → 物理 sim2real) · Paper C (WM-Guided Diff+WMID → 算法侧重)"},
		{"id":"breakthrough_group","type":"group","x":-1260,"y":-700,"width":2920,"height":550,"color":"2","label":"⚡ 算法突破点 (Key Technical Challenges)"},
		{"id":"bt_frequency","type":"text","x":-1210,"y":-640,"width":350,"height":320,"color":"2","text":"## 🕐 控制频率困境\n\n**痛点**: 仿真高频 vs 真机低频\n\n**信号/控制约束**:\n- Nyquist采样与混叠\n- Bode带宽/相位裕度\n- 滤波与通信延迟\n\n**解决思路**:\n- 动作持续时间学习\n- 弹性时间步 (VTS)\n- 惯性阶段→自动低频\n- 接触切换→自动高频\n\n→ [[ControlTheory]] [[SignalProcessing]] [[ReinforcementLearning]]"},
		{"id":"bt_sparse","type":"text","x":-810,"y":-640,"width":350,"height":280,"color":"2","text":"## 🔍 稀疏奖励探索\n\n**痛点**: 长因果链导致奖励稀疏\n\n**解决思路**:\n- 目标重标注 (HER)\n- 速度缩放拉开惯性窗口\n- 特权动作简化探索\n- 隐式课程学习\n\n→ [[ReinforcementLearning]] [[InformationTheory]]"},
		{"id":"bt_impedance","type":"text","x":-410,"y":-640,"width":350,"height":350,"color":"2","text":"## 🎛️ 变阻抗控制\n\n**痛点**: 固定PD低频僵硬\n\n**解决思路**:\n- 刚度作为动作空间维度\n- 接触时低刚度顺从\n- 运动时高刚度精确\n- 末端空间参数化\n\n📊 **Exp2验证**: Kp最优3.5~8.5\n灵敏度极高→支持变阻抗\n\n→ [[ControlTheory]] [[ContactMechanics]]"},
		{"id":"bt_curriculum","type":"text","x":-10,"y":-640,"width":350,"height":350,"color":"2","text":"## 📚 物理参数课程\n\n**痛点**: 直接学习高动态任务困难\n\n**解决思路**:\n- α-scaling 速度缩放\n- 从简单到复杂分布\n- 重力/摩擦渐进\n- Continuation Method\n\n⚡ **Exp2发现**: TWC对TP有效\n但对TA无效! → 需任务特化课程\n\n→ [[Optimization]] [[ReinforcementLearning]]"},
		{"id":"bt_sim2real","type":"text","x":390,"y":-640,"width":350,"height":350,"color":"2","text":"## 🌐 Sim-to-Real 迁移\n\n**痛点**: 动力学域差异\n\n**解决思路**:\n- Real-to-Sim-to-Real\n- 在线校正 (TRANSIC)\n- 数字孪生 RL 鲁棒化\n- 域随机化 + 快速电机适应\n- 🔧 电机/减速器/传动建模\n- ⏱️ CAN latency + 指间相位差\n- 🧠 System ID + 神经动力学\n- 📐 数据驱动 LMI 稳定证书\n\n**MDP四要素Gap分类**:\nState | Action | Transition | Reward\n\n→ [[ReinforcementLearning]] [[Dynamics]] [[ControlTheory]]\n→ [[sim2real|硬件Gap分析]]"},
		{"id":"bt_privileged","type":"text","x":790,"y":-640,"width":350,"height":280,"color":"2","text":"## 🔮 特权动作\n\n**痛点**: 物理边界阻碍探索\n\n**解决思路**:\n- 禁用碰撞穿透学习\n- 虚拟力辅助\n- 课程恢复物理约束\n- 从作弊到真实\n\n→ [[ContactMechanics]] [[ReinforcementLearning]]"},
		{"id":"bt_representation","type":"text","x":1190,"y":-640,"width":350,"height":280,"color":"2","text":"## 🧩 感知与表征\n\n**痛点**: 高维观测空间\n\n**解决思路**:\n- 视触觉融合\n- 扩散策略分布建模\n- 点云几何特征\n- 物理属性嵌入\n\n→ [[RepresentationLearning]] [[SignalProcessing]]"},
		{"id":"papers_group","type":"group","x":-1260,"y":30,"width":3300,"height":1830,"color":"4","label":"📄 PapersRecap: 核心算法与洞见"},
		{"id":"paper_tarc","type":"text","x":-1210,"y":90,"width":350,"height":200,"color":"5","text":"## TARC (2025)\n**时间自适应控制**\n\n💡 策略输出 (a, Δt)\n惯性阶段→低频\n接触切换→高频\n\n⭐ 直接解决频率困境"},
		{"id":"paper_pfqi","type":"text","x":-1210,"y":340,"width":350,"height":200,"color":"5","text":"## Action Persistence (2020)\n**动作持续理论框架**\n\n💡 k-persistent MDP\n性能损失 ∝ 环境演化速度\n\n⭐ 频率选择的理论基础"},
		{"id":"paper_multifreq","type":"text","x":-1210,"y":590,"width":350,"height":200,"color":"5","text":"## Multi-Frequency RL (2020)\n**多频率控制理论**\n\n💡 高层低频决策\n底层高频跟踪\n层次化频率分配\n\n⭐ 频率分离框架"},
		{"id":"paper_elastic","type":"text","x":-1210,"y":840,"width":350,"height":180,"color":"5","text":"## VTS-RL (2021)\n**弹性时间步**\n\n💡 时间缩放MDP\n动态调整Δt\n\n⭐ TARC前身"},
		{"id":"paper_her","type":"text","x":-810,"y":90,"width":350,"height":200,"color":"4","text":"## HER (2017)\n**Hindsight Experience Replay**\n\n💡 失败→成功\n目标重标注\n隐式课程学习\n\n⭐ 稀疏奖励基石"},
		{"id":"paper_eureka","type":"text","x":-810,"y":340,"width":350,"height":200,"color":"4","text":"## EUREKA (2023)\n**LLM自动奖励设计**\n\n💡 LLM 生成奖励函数\n无需人工设计\n超越人类级别\n\n⭐ 自动化奖励工程"},
		{"id":"paper_vices","type":"text","x":-410,"y":90,"width":350,"height":200,"color":"4","text":"## VICES (2019)\n**末端空间变阻抗**\n\n💡 动作 = (Δx, K)\n位移 + 刚度\n任务相关顺从性\n\n⭐ 接触任务动作空间"},
		{"id":"paper_facet","type":"text","x":-410,"y":340,"width":350,"height":200,"color":"4","text":"## FACET (2025)\n**阻抗参考模型跟踪**\n\n💡 RL跟踪虚拟弹簧-质量-阻尼\n统一接口 (x_des, Kp, Kd)\n时变阻抗自然匹配相位\n\n⭐ DNPM 方向A核心方案"},
		{"id":"paper_lipsnet","type":"text","x":-410,"y":590,"width":350,"height":200,"color":"4","text":"## LipsNet (2023)\n**自适应Lipschitz神经网络**\n\n💡 自适应 Lipschitz 约束\n平滑策略输出\n鲁棒最优控制\n\n⭐ Idea-006 ALA 理论来源"},
		{"id":"paper_curriculum","type":"text","x":-10,"y":90,"width":350,"height":200,"color":"4","text":"## Curriculum Learning (2009)\n**课程学习理论**\n\n💡 Continuation Method\n简单→复杂分布\n避免局部极小\n\n⭐ α-scaling 理论背书"},
		{"id":"paper_demostart","type":"text","x":-10,"y":340,"width":350,"height":200,"color":"4","text":"## DemoStart (2024)\n**演示引导自动课程**\n\n💡 从人类演示起始\n自动难度递增\n多指灵巧操作\n\n⭐ 课程+Sim2Real 融合"},
		{"id":"paper_rialto","type":"text","x":390,"y":90,"width":350,"height":200,"color":"5","text":"## RialTo (2024)\n**Real-to-Sim-to-Real**\n\n💡 数字孪生+RL鲁棒化\n逆向蒸馏\n67%+ 鲁棒性提升\n\n⭐ 迁移新范式"},
		{"id":"paper_pen","type":"text","x":390,"y":340,"width":350,"height":200,"color":"5","text":"## Pen Spinning (2024)\n**转笔三阶段Sim2Real**\n\n💡 Oracle + OpenLoop + Finetune\n<50条真实轨迹\nFinger Gaiting\n\n⭐ 动态操作典范"},
		{"id":"paper_transic","type":"text","x":390,"y":590,"width":350,"height":200,"color":"5","text":"## TRANSIC (2024)\n**在线校正迁移**\n\n💡 从在线人类校正学习\n残差策略\n无需额外仿真\n\n⭐ 迁移Gap动态修复"},
		{"id":"paper_hora","type":"text","x":390,"y":840,"width":350,"height":200,"color":"5","text":"## HORA (2022)\n**快速电机适应**\n\n💡 Rapid Motor Adaptation\n在线系统辨识\n隐式参数推断\n\n⭐ 零样本域适应"},
		{"id":"paper_privileged","type":"text","x":790,"y":90,"width":350,"height":200,"color":"4","text":"## Privileged Action (2025)\n**特权动作课程学习**\n\n💡 仿真中作弊\n禁用碰撞/虚拟力\n课程恢复约束\n\n⭐ 长时程探索突破"},
		{"id":"paper_longhorizon","type":"text","x":790,"y":340,"width":350,"height":200,"color":"4","text":"## Long-Horizon (2025)\n**特权动作长时程**\n\n💡 Push-and-Grasp\nPivot Grasp\n行为自发涌现\n\n⭐ 无需手工子任务"},
		{"id":"paper_rotateit","type":"text","x":1190,"y":90,"width":350,"height":200,"color":"5","text":"## RotateIt (2024)\n**视触觉旋转**\n\n💡 Vision + Touch融合\n通用物体旋转\n多模态表征\n\n⭐ 表征学习典范"},
		{"id":"paper_anyrotate","type":"text","x":1190,"y":340,"width":350,"height":200,"color":"5","text":"## AnyRotate (2024)\n**重力不变旋转**\n\n💡 任意方向手内旋转\nSim-to-Real Touch\n触觉驱动策略\n\n⭐ 触觉闭环操作"},
		{"id":"paper_hato","type":"text","x":1190,"y":590,"width":350,"height":200,"color":"5","text":"## HATO (2024)\n**双手视触觉操作**\n\n💡 两只多指手协作\n视触觉技能学习\n多模态感知融合\n\n⭐ 双手灵巧操作"},
		{"id":"foundations_group","type":"group","x":-1260,"y":1400,"width":2920,"height":840,"color":"6","label":"🧠 Foundations: 理论基础"},
		{"id":"found_rl","type":"file","file":"Foundations/ReinforcementLearning.md","x":-1210,"y":1460,"width":350,"height":120},
		{"id":"found_rl_note","type":"text","x":-1210,"y":1600,"width":350,"height":100,"color":"6","text":"**核心**: TD→PG→PPO/SAC · 接触流形探索 · Sim-to-Real"},
		{"id":"found_control","type":"file","file":"Foundations/ControlTheory.md","x":-810,"y":1460,"width":350,"height":120},
		{"id":"found_control_note","type":"text","x":-810,"y":1600,"width":350,"height":120,"color":"6","text":"**核心**: 传递函数/状态空间 · PID→CTC→阻抗/导纳 · LQR · 数据驱动LMI证书"},
		{"id":"found_dynamics","type":"file","file":"Foundations/Dynamics.md","x":-410,"y":1460,"width":350,"height":120},
		{"id":"found_dynamics_note","type":"text","x":-410,"y":1600,"width":350,"height":100,"color":"6","text":"**核心**: Lagrangian/Hamiltonian · 参数线性性 (Y·π=τ) · RNEA/ABA · 接触动力学"},
		{"id":"found_contact","type":"file","file":"Foundations/ContactMechanics.md","x":-10,"y":1460,"width":350,"height":120},
		{"id":"found_contact_note","type":"text","x":-10,"y":1600,"width":350,"height":100,"color":"6","text":"**核心**: 摩擦锥约束 · 接触切换 · 力学交互"},
		{"id":"found_optim","type":"file","file":"Foundations/Optimization.md","x":390,"y":1460,"width":350,"height":120},
		{"id":"found_optim_note","type":"text","x":390,"y":1600,"width":350,"height":100,"color":"6","text":"**核心**: Continuation · iLQR/MPC · 非凸优化"},
		{"id":"found_signal","type":"file","file":"Foundations/SignalProcessing.md","x":790,"y":1460,"width":350,"height":120},
		{"id":"found_signal_note","type":"text","x":790,"y":1600,"width":350,"height":120,"color":"6","text":"**核心**: Fourier/STFT/Wavelet · 采样/滤波 · 触觉状态估计 · KF→EKF→UKF→PF"},
		{"id":"found_stochastic","type":"file","file":"Foundations/StochasticProcess.md","x":-810,"y":1750,"width":350,"height":120},
		{"id":"found_stochastic_note","type":"text","x":-810,"y":1890,"width":350,"height":100,"color":"6","text":"**核心**: 扩散策略 · 不确定性建模 · MPPI采样优化"},
		{"id":"found_repr","type":"file","file":"Foundations/RepresentationLearning.md","x":-410,"y":1750,"width":350,"height":120},
		{"id":"found_repr_note","type":"text","x":-410,"y":1890,"width":350,"height":100,"color":"6","text":"**核心**: PCA→AE→VAE→对比 · 视触觉融合 · 点云"},
		{"id":"found_info","type":"file","file":"Foundations/InformationTheory.md","x":-10,"y":1750,"width":350,"height":120},
		{"id":"found_info_note","type":"text","x":-10,"y":1890,"width":350,"height":100,"color":"6","text":"**核心**: 率失真 · 压缩-去噪对偶性 · 信息瓶颈"},
		{"id":"found_compgeo","type":"file","file":"Foundations/ComputationalGeometry.md","x":390,"y":1750,"width":350,"height":120},
		{"id":"found_compgeo_note","type":"text","x":390,"y":1890,"width":350,"height":100,"color":"6","text":"**核心**: SDF · 碰撞检测 · 神经隐式表示"},
		{"id":"found_embodied","type":"file","file":"Foundations/EmbodiedAI.md","x":790,"y":1750,"width":350,"height":120},
		{"id":"found_embodied_note","type":"text","x":790,"y":1890,"width":350,"height":100,"color":"6","text":"**核心**: VLA模型 · 仿真器生态 · 端到端策略"},
		{"id":"paper_rl100","type":"text","x":-1210,"y":1100,"width":350,"height":200,"color":"4","text":"## RL-100 (2025)\n**Denoising Sub-MDP**\n\n💡 扩散策略的RL微调\nIL→Offline→Online RL\nConsistency Distillation\n\n⭐ 7任务 100% SR"},
		{"id":"paper_wmpo","type":"text","x":-810,"y":1100,"width":350,"height":200,"color":"4","text":"## WMPO (2025)\n**世界模型策略优化**\n\n💡 像素空间世界模型\nGRPO + VLM-as-Judge\nVLA RL Post-Training\n\n⭐ 动态采样策略"},
		{"id":"paper_last0","type":"text","x":-410,"y":1100,"width":350,"height":200,"color":"4","text":"## LaST0 (2025)\n**潜在时空CoT VLA**\n\n💡 Mixture-of-Thought\n快慢双系统路由\n14× 推理加速\n\n⭐ 频率困境新视角"},
		{"id":"paper_omnix","type":"text","x":-10,"y":1100,"width":350,"height":200,"color":"5","text":"## OmniXtreme (2025)\n**高动态人形控制**\n\n💡 Flow Matching预训练\nActuation-aware残差RL\nTorque-speed envelope\n\n⭐ Sim2Real执行器建模"},
		{"id":"paper_geopt","type":"text","x":390,"y":1100,"width":350,"height":200,"color":"4","text":"## GeoPT (2025)\n**Dynamics-Lifted预训练**\n\n💡 Transport equation\nE(3)-equivariant\n数据需求↓20-60%\n\n⭐ 物理感知表征"},
		{"id":"paper_cgp","type":"text","x":790,"y":1100,"width":350,"height":210,"text":"## CGP (2026)\n**接触Grounding扩散策略**\n\n💡 耦合状态-触觉扩散\n接触一致性映射 Mφ\n潜在触觉VAE\n\n⭐ 视触觉灵巧操作"},
		{"id":"paper_mcc","type":"text","x":1190,"y":1100,"width":350,"height":210,"text":"## MCC (2026)\n**无传感器柔顺控制**\n\n💡 电机电流估计接触力\n方向相关效率模型\n弹簧-质量-阻尼导纳\n\n⭐ 高减速比伺服适用"},
		{"id":"paper_dexhil","type":"text","x":-1210,"y":1350,"width":350,"height":210,"text":"## DexHiL (2026)\n**HiL VLA后训练**\n\n💡 首个臂手HiL VLA\n干预感知采样\nDAgger循环\n+25% 成功率\n\n⭐ VLA实时修正"},
		{"id":"paper_tacmap","type":"text","x":-810,"y":1350,"width":350,"height":210,"text":"## Tacmap (2026)\n**穿透深度图触觉**\n\n💡 穿透深度=统一触觉表征\n几何无关(曲面指尖)\n零样本sim-to-real\n\n⭐ 触觉Sim2Real"},
		{"id":"paper_dapl","type":"text","x":-410,"y":1350,"width":350,"height":210,"text":"## DAPL (2026)\n**动力学感知外在灵巧**\n\n💡 世界模型预测接触动力学\n动力学条件化RL\n杂乱场景推、滑、翻\n\n⭐ 非紧握操作"},
		{"id":"paper_gat","type":"text","x":-10,"y":1350,"width":350,"height":210,"text":"## GAT (2017)\n**Grounded Action变换**\n\n💡 学习动作映射修正sim\nGSL迭代grounding\nNAO行走+43%\n\n⭐ Sim2Real经典"},
		{"id":"paper_stola","type":"text","x":390,"y":1350,"width":350,"height":210,"text":"## SToLa (2026)\n**触觉-语言MoE**\n\n💡 MoE动态路由\n触觉/语言专家分配\n开放场景常识推理\n\n⭐ 触觉理解"},
		{"id":"paper_robotwin2","type":"text","x":790,"y":1350,"width":350,"height":210,"text":"## RoboTwin 2.0 (2025)\n**双臂数据生成**\n\n💡 MLLM代码生成\n5轴域随机化\n731物体50任务\n零样本+228%\n\n⭐ 数据工厂"},
		{"id":"paper_pointworld","type":"text","x":1190,"y":1350,"width":350,"height":210,"text":"## PointWorld (2026)\n**空间智能3D表征**\n\n💡 3D Flow统一动作空间\n载体无关世界模型\n迁移效率比NLP低100×\n\n⭐ 世界模型核心"},
		{"id":"paper_sim2real_survey","type":"text","x":-1210,"y":1600,"width":350,"height":210,"text":"## Sim2Real Survey (2025)\n**MDP四要素分类框架**\n\n💡 State/Action/Transition/Reward\nFoundation Model赋能\nAwesomeSim2Real\n\n⭐ 综述"},
		{"id":"paper_sim2real_review","type":"text","x":-810,"y":1600,"width":350,"height":210,"text":"## Sim2Real Review (2026)\n**执行器级建模视角**\n\n💡 仿真逼真度+域随机化\n执行器非线性\n多保真度学习\n\n⭐ 综述"},
		{"id":"paper_reducers","type":"text","x":-410,"y":1600,"width":350,"height":210,"text":"## 谐波vs RV减速器 (2026)\n**关节传动选型**\n\n💡 谐波:轻载紧凑空心轴\nRV:重载高刚度\n7维决策矩阵\n\n⭐ 硬件基础"},
		{"id":"proj_mech_hw","type":"text","x":750,"y":-1640,"width":400,"height":540,"color":"1","text":"## 🔧 灵巧手机械结构\n\n**传动**: 腱绳 · 连杆 · 直驱 · QDD\n**电机**: BLDC(FOC) · 空心杯 · 伺服\n**减速器**: 谐波 · 行星 · RV · 摆线\n\n**Sim-to-Real 核心 Gap**:\n→ 电气时间常数被忽略\n→ 齿隙/非线性摩擦\n→ 欠驱动耦合矩阵 R(q)\n→ 温度漂移\n\n→ [[sim2real|硬件Gap分析]]\n→ [[传动]] [[电机]] [[减速器]]"},
		{"id":"paper_dexndm","type":"text","x":-10,"y":1600,"width":350,"height":210,"text":"## DexNDM (2025)\n**关节级神经动力学**\n\n💡 Joint-wise Neural Dynamics\n残差补偿Sim-Real Gap\n在线适应 (无需DR)\n\n⭐ System ID前沿"},
		{"id":"paper_wog","type":"text","x":390,"y":1600,"width":350,"height":210,"color":"4","text":"## WoG (2026)\n**条件空间世界建模**\n\n💡 未来观测→紧凑条件空间\n2阶段: Guidance→Inference\nQ-Former+DiT Rectified Flow\n\n⭐ VLA世界模型新范式"},
		{"id":"paper_comet","type":"text","x":790,"y":1600,"width":350,"height":210,"text":"## COMET (2025)\n**可控长时域运动生成**\n\n💡 Transformer CVAE自回归\nGMM参考引导反馈\n任意关节子集控制\n\n⭐ 运动生成→灵巧操作迁移"},
		{"id":"paper_phygile","type":"text","x":1190,"y":1600,"width":350,"height":210,"color":"5","text":"## PhyGile (2026)\n**Physics-Prefix运动生成**\n\n💡 课程MoE跟踪 + 262D扩散\n物理前缀引导去噪\nTP-MoE细粒度对齐\n\n⭐ 闭环生成-执行"},
		{"id":"paper_rlt","type":"text","x":1590,"y":1100,"width":350,"height":210,"color":"5","text":"## RLT (2026)\n**RL Token在线精细化**\n\n💡 VLA→RL Token信息瓶颈\n轻量actor-critic本地训练\n残差动作编辑 15min适应\n\n⭐ VLA最后一毫米精度"},
		{"id":"paper_act","type":"text","x":1590,"y":90,"width":350,"height":200,"color":"4","text":"## ACT (2023)\n**动作分块Transformer**\n\n💡 CVAE + Temporal Ensembling\n低成本双臂精细操作\n端到端模仿学习\n\n⭐ 双臂操作基石"},
		{"id":"paper_recap","type":"text","x":1590,"y":340,"width":350,"height":200,"color":"4","text":"## RECAP / π₀.6 (2025)\n**经验驱动VLA后训练**\n\n💡 三阶段: IL→Offline→Online RL\n优势条件化策略\nAWR/GRPO统一视角\n\n⭐ VLA RL Post-Training"},
		{"id":"paper_unified","type":"text","x":1590,"y":590,"width":350,"height":200,"color":"4","text":"## Unified Policy (2024)\n**On/Off-Policy统一框架**\n\n💡 Eval/Improve二维分解\nUpdate Schedule三态分类\nPPO/SAC/BRAC统一推导\n\n⭐ RL算法分类基石"},
		{"id":"proj_meeting_pipeline","type":"file","file":"Projects/Dynamic Non-Prehensile Manipulation/Dynamic Non-Prehensile Manipulation.md","subpath":"#6.5 Meeting-Synthesized Research Pipeline（2026-03 会议综合）","x":640,"y":-1300,"width":520,"height":220},
		{"id":"found_theory_spine","type":"text","x":-1210,"y":2020,"width":2350,"height":160,"color":"6","text":"## Foundation 理论大厦骨架\n\n每个 Foundation 已补入 §0 路线：物理/数学起源 → 算法实现 → 灵巧操作失败模式 → 项目映射。\n\n主链: Dynamics/Contact/Geometry → Control/Optimization → Signal/Stochastic/Info → Representation/RL/EmbodiedAI\n\n→ [[Foundations/taxonomy#Foundation 理论大厦骨架索引]]"}
	],
	"edges":[
		{"id":"e_exp2_i001","fromNode":"exp2_findings","fromSide":"right","toNode":"idea_001","toSide":"left","color":"5","label":"Kp+基线"},
		{"id":"e_exp2_i003","fromNode":"exp2_findings","fromSide":"bottom","toNode":"idea_003","toSide":"top","color":"5","label":"reward hacking"},
		{"id":"e_exp2_i006","fromNode":"exp2_findings","fromSide":"bottom","toNode":"idea_006","toSide":"top","color":"5","label":"Heavy失败"},
		{"id":"e_exp2_i007","fromNode":"exp2_findings","fromSide":"right","toNode":"idea_007","toSide":"left","color":"5","label":"TWC不对称"},
		{"id":"e_i001_proj","fromNode":"idea_001","fromSide":"bottom","toNode":"proj_dnpm_insight","toSide":"top","color":"1","label":"PAI"},
		{"id":"e_i002_proj","fromNode":"idea_002","fromSide":"bottom","toNode":"proj_dnpm_insight","toSide":"top","color":"1","label":"CA-ARP"},
		{"id":"e_i007_proj","fromNode":"idea_007","fromSide":"bottom","toNode":"proj_dnpm_insight","toSide":"top","color":"1","label":"DOC"},
		{"id":"e_i006_proj","fromNode":"idea_006","fromSide":"bottom","toNode":"proj_dnpm_insight","toSide":"top","color":"1","label":"ALA"},
		{"id":"e_i003_proj","fromNode":"idea_003","fromSide":"bottom","toNode":"proj_dnpm_insight","toSide":"top","color":"1","label":"CMR"},
		{"id":"e_proj_bt1","fromNode":"proj_dnpm_insight","fromSide":"bottom","toNode":"bt_frequency","toSide":"top","label":"痛点1"},
		{"id":"e_proj_bt2","fromNode":"proj_dnpm_insight","fromSide":"bottom","toNode":"bt_sparse","toSide":"top","label":"痛点2"},
		{"id":"e_proj_bt3","fromNode":"proj_dnpm_insight","fromSide":"bottom","toNode":"bt_impedance","toSide":"top","label":"痛点3"},
		{"id":"e_proj_bt4","fromNode":"proj_dnpm_insight","fromSide":"bottom","toNode":"bt_curriculum","toSide":"top","label":"痛点4"},
		{"id":"e_proj_bt5","fromNode":"proj_dnpm_insight","fromSide":"bottom","toNode":"bt_sim2real","toSide":"top","label":"痛点5"},
		{"id":"e_wmts_bt5","fromNode":"proj_wmts_insight","fromSide":"bottom","toNode":"bt_sim2real","toSide":"right","color":"1","label":"Actuator建模"},
		{"id":"e_wmts_dnpm","fromNode":"proj_dnpm_insight","fromSide":"right","toNode":"proj_wmts_core","toSide":"left","color":"1","label":"WM调度"},
		{"id":"e_wmts_reliability_core","fromNode":"proj_wmts_core","fromSide":"right","toNode":"proj_wmts_reliability","toSide":"left","color":"1","label":"可靠性增强"},
		{"id":"e_wmts_reliability_papers","fromNode":"proj_wmts_papers","fromSide":"right","toNode":"proj_wmts_reliability","toSide":"left","color":"4","label":"P0论文支撑"},
		{"id":"e_wmts_reliability_sim2real","fromNode":"proj_wmts_reliability","fromSide":"bottom","toNode":"bt_sim2real","toSide":"top","color":"1","label":"保守放行"},
		{"id":"e_wmts_insights_core","fromNode":"proj_wmts_insights_index","fromSide":"left","toNode":"proj_wmts_core","toSide":"right","color":"4","label":"15 真机 RL idea"},
		{"id":"e_wmts_insights_reliability","fromNode":"proj_wmts_insights_index","fromSide":"top","toNode":"proj_wmts_reliability","toSide":"bottom","color":"4","label":"互补主线"},
		{"id":"e_wmts_insights_summary_index","fromNode":"proj_wmts_insights_summary","fromSide":"top","toNode":"proj_wmts_insights_index","toSide":"bottom","color":"4"},
		{"id":"e_wmts_insights_sim2real","fromNode":"proj_wmts_insights_summary","fromSide":"bottom","toNode":"bt_sim2real","toSide":"top","color":"4","label":"sim2real 修复"},
		{"id":"e_proj_bt6","fromNode":"proj_dnpm_insight","fromSide":"bottom","toNode":"bt_privileged","toSide":"top","label":"痛点6"},
		{"id":"e_proj_bt7","fromNode":"proj_dnpm_insight","fromSide":"bottom","toNode":"bt_representation","toSide":"top","label":"痛点7"},
		{"id":"e_i001_bt3","fromNode":"idea_001","fromSide":"bottom","toNode":"bt_impedance","toSide":"top","color":"3","label":"核心"},
		{"id":"e_i002_bt2","fromNode":"idea_002","fromSide":"bottom","toNode":"bt_sparse","toSide":"top","color":"3"},
		{"id":"e_i007_bt4","fromNode":"idea_007","fromSide":"bottom","toNode":"bt_curriculum","toSide":"top","color":"3","label":"核心"},
		{"id":"e_i006_bt5","fromNode":"idea_006","fromSide":"bottom","toNode":"bt_sim2real","toSide":"top","color":"3","label":"核心"},
		{"id":"e_i003_bt2","fromNode":"idea_003","fromSide":"bottom","toNode":"bt_sparse","toSide":"top","color":"3","label":"核心"},
		{"id":"e_i005_bt5","fromNode":"idea_005","fromSide":"bottom","toNode":"bt_sim2real","toSide":"top","color":"3","label":"核心"},
		{"id":"e_bt1_tarc","fromNode":"bt_frequency","fromSide":"bottom","toNode":"paper_tarc","toSide":"top","color":"5","label":"⭐"},
		{"id":"e_bt1_pfqi","fromNode":"bt_frequency","fromSide":"bottom","toNode":"paper_pfqi","toSide":"top"},
		{"id":"e_bt1_multi","fromNode":"bt_frequency","fromSide":"bottom","toNode":"paper_multifreq","toSide":"top"},
		{"id":"e_bt1_vts","fromNode":"bt_frequency","fromSide":"bottom","toNode":"paper_elastic","toSide":"top"},
		{"id":"e_bt2_her","fromNode":"bt_sparse","fromSide":"bottom","toNode":"paper_her","toSide":"top","color":"4","label":"⭐"},
		{"id":"e_bt2_eureka","fromNode":"bt_sparse","fromSide":"bottom","toNode":"paper_eureka","toSide":"top"},
		{"id":"e_bt3_vices","fromNode":"bt_impedance","fromSide":"bottom","toNode":"paper_vices","toSide":"top","color":"4","label":"⭐"},
		{"id":"e_bt3_facet","fromNode":"bt_impedance","fromSide":"bottom","toNode":"paper_facet","toSide":"top","color":"4","label":"⭐"},
		{"id":"e_bt3_lips","fromNode":"bt_impedance","fromSide":"bottom","toNode":"paper_lipsnet","toSide":"top"},
		{"id":"e_bt4_curri","fromNode":"bt_curriculum","fromSide":"bottom","toNode":"paper_curriculum","toSide":"top","color":"4","label":"⭐"},
		{"id":"e_bt4_demo","fromNode":"bt_curriculum","fromSide":"bottom","toNode":"paper_demostart","toSide":"top"},
		{"id":"e_bt5_rial","fromNode":"bt_sim2real","fromSide":"bottom","toNode":"paper_rialto","toSide":"top","color":"5","label":"⭐"},
		{"id":"e_bt5_pen","fromNode":"bt_sim2real","fromSide":"bottom","toNode":"paper_pen","toSide":"top"},
		{"id":"e_bt5_tran","fromNode":"bt_sim2real","fromSide":"bottom","toNode":"paper_transic","toSide":"top"},
		{"id":"e_bt5_hora","fromNode":"bt_sim2real","fromSide":"bottom","toNode":"paper_hora","toSide":"top"},
		{"id":"e_bt6_priv","fromNode":"bt_privileged","fromSide":"bottom","toNode":"paper_privileged","toSide":"top","color":"4","label":"⭐"},
		{"id":"e_bt6_long","fromNode":"bt_privileged","fromSide":"bottom","toNode":"paper_longhorizon","toSide":"top"},
		{"id":"e_bt7_rot","fromNode":"bt_representation","fromSide":"bottom","toNode":"paper_rotateit","toSide":"top","color":"5","label":"⭐"},
		{"id":"e_bt7_any","fromNode":"bt_representation","fromSide":"bottom","toNode":"paper_anyrotate","toSide":"top"},
		{"id":"e_bt7_hato","fromNode":"bt_representation","fromSide":"bottom","toNode":"paper_hato","toSide":"top"},
		{"id":"e_tarc_rl","fromNode":"paper_tarc","fromSide":"bottom","toNode":"found_rl","toSide":"top","color":"6"},
		{"id":"e_tarc_ctrl","fromNode":"paper_tarc","fromSide":"bottom","toNode":"found_control","toSide":"top","color":"6"},
		{"id":"e_pfqi_rl","fromNode":"paper_pfqi","fromSide":"bottom","toNode":"found_rl","toSide":"top","color":"6"},
		{"id":"e_her_rl","fromNode":"paper_her","fromSide":"bottom","toNode":"found_rl","toSide":"top","color":"6"},
		{"id":"e_vices_ctrl","fromNode":"paper_vices","fromSide":"bottom","toNode":"found_control","toSide":"top","color":"6"},
		{"id":"e_vices_cnt","fromNode":"paper_vices","fromSide":"bottom","toNode":"found_contact","toSide":"top","color":"6"},
		{"id":"e_facet_ctrl","fromNode":"paper_facet","fromSide":"bottom","toNode":"found_control","toSide":"top","color":"6"},
		{"id":"e_facet_dyn","fromNode":"paper_facet","fromSide":"bottom","toNode":"found_dynamics","toSide":"top","color":"6"},
		{"id":"e_lips_ctrl","fromNode":"paper_lipsnet","fromSide":"bottom","toNode":"found_control","toSide":"top","color":"6"},
		{"id":"e_curri_opt","fromNode":"paper_curriculum","fromSide":"bottom","toNode":"found_optim","toSide":"top","color":"6"},
		{"id":"e_demo_rl","fromNode":"paper_demostart","fromSide":"bottom","toNode":"found_rl","toSide":"top","color":"6"},
		{"id":"e_rial_rl","fromNode":"paper_rialto","fromSide":"bottom","toNode":"found_rl","toSide":"top","color":"6"},
		{"id":"e_pen_dyn","fromNode":"paper_pen","fromSide":"bottom","toNode":"found_dynamics","toSide":"top","color":"6"},
		{"id":"e_pen_cnt","fromNode":"paper_pen","fromSide":"bottom","toNode":"found_contact","toSide":"top","color":"6"},
		{"id":"e_tran_rl","fromNode":"paper_transic","fromSide":"bottom","toNode":"found_rl","toSide":"top","color":"6"},
		{"id":"e_hora_rl","fromNode":"paper_hora","fromSide":"bottom","toNode":"found_rl","toSide":"top","color":"6"},
		{"id":"e_hora_dyn","fromNode":"paper_hora","fromSide":"bottom","toNode":"found_dynamics","toSide":"top","color":"6"},
		{"id":"e_priv_cnt","fromNode":"paper_privileged","fromSide":"bottom","toNode":"found_contact","toSide":"top","color":"6"},
		{"id":"e_long_cnt","fromNode":"paper_longhorizon","fromSide":"bottom","toNode":"found_contact","toSide":"top","color":"6"},
		{"id":"e_rot_repr","fromNode":"paper_rotateit","fromSide":"bottom","toNode":"found_repr","toSide":"top","color":"6"},
		{"id":"e_rot_sig","fromNode":"paper_rotateit","fromSide":"bottom","toNode":"found_signal","toSide":"top","color":"6"},
		{"id":"e_any_sig","fromNode":"paper_anyrotate","fromSide":"bottom","toNode":"found_signal","toSide":"top","color":"6"},
		{"id":"e_hato_repr","fromNode":"paper_hato","fromSide":"bottom","toNode":"found_repr","toSide":"top","color":"6"},
		{"id":"e_eureka_rl","fromNode":"paper_eureka","fromSide":"bottom","toNode":"found_rl","toSide":"top","color":"6"},
		{"id":"e_vts_tarc","fromNode":"paper_elastic","fromSide":"right","toNode":"paper_tarc","toSide":"bottom","color":"3","label":"演进"},
		{"id":"e_vices_facet","fromNode":"paper_vices","fromSide":"bottom","toNode":"paper_facet","toSide":"top","color":"3","label":"演进"},
		{"id":"e_curri_demo","fromNode":"paper_curriculum","fromSide":"bottom","toNode":"paper_demostart","toSide":"top","color":"3","label":"实践化"},
		{"id":"e_rial_pen","fromNode":"paper_rialto","fromSide":"bottom","toNode":"paper_pen","toSide":"top","color":"3","label":"互补范式"},
		{"id":"e_priv_long","fromNode":"paper_privileged","fromSide":"bottom","toNode":"paper_longhorizon","toSide":"top","color":"3","label":"应用拓展"},
		{"id":"e_hora_pen","fromNode":"paper_hora","fromSide":"left","toNode":"paper_pen","toSide":"right","color":"3","label":"基础框架"},
		{"id":"e_curri_her","fromNode":"paper_curriculum","fromSide":"left","toNode":"paper_her","toSide":"right","color":"3","label":"隐式课程"},
		{"id":"e_bt2_rl100","fromNode":"bt_sparse","fromSide":"bottom","toNode":"paper_rl100","toSide":"top","color":"4"},
		{"id":"e_rl100_rl","fromNode":"paper_rl100","fromSide":"bottom","toNode":"found_rl","toSide":"top"},
		{"id":"e_rl100_stoch","fromNode":"paper_rl100","fromSide":"bottom","toNode":"found_stochastic","toSide":"top"},
		{"id":"e_bt2_wmpo","fromNode":"bt_sparse","fromSide":"bottom","toNode":"paper_wmpo","toSide":"top","color":"4","label":"动态采样"},
		{"id":"e_wmpo_rl","fromNode":"paper_wmpo","fromSide":"bottom","toNode":"found_rl","toSide":"top"},
		{"id":"e_wmpo_embodied","fromNode":"paper_wmpo","fromSide":"bottom","toNode":"found_embodied","toSide":"top"},
		{"id":"e_bt1_last0","fromNode":"bt_frequency","fromSide":"bottom","toNode":"paper_last0","toSide":"top","color":"4","label":"快慢系统"},
		{"id":"e_last0_embodied","fromNode":"paper_last0","fromSide":"bottom","toNode":"found_embodied","toSide":"top"},
		{"id":"e_last0_repr","fromNode":"paper_last0","fromSide":"bottom","toNode":"found_repr","toSide":"top"},
		{"id":"e_bt5_omnix","fromNode":"bt_sim2real","fromSide":"bottom","toNode":"paper_omnix","toSide":"top","color":"5","label":"执行器感知"},
		{"id":"e_omnix_control","fromNode":"paper_omnix","fromSide":"bottom","toNode":"found_control","toSide":"top"},
		{"id":"e_omnix_dynamics","fromNode":"paper_omnix","fromSide":"bottom","toNode":"found_dynamics","toSide":"top"},
		{"id":"e_bt7_geopt","fromNode":"bt_representation","fromSide":"bottom","toNode":"paper_geopt","toSide":"top","color":"4","label":"物理表征"},
		{"id":"e_geopt_dynamics","fromNode":"paper_geopt","fromSide":"bottom","toNode":"found_dynamics","toSide":"top"},
		{"id":"e_geopt_compgeo","fromNode":"paper_geopt","fromSide":"bottom","toNode":"found_compgeo","toSide":"top"},
		{"id":"e_rl100_wmpo","fromNode":"paper_rl100","fromSide":"right","toNode":"paper_wmpo","toSide":"left","label":"互补: 真实vs世界模型"},
		{"id":"a6107b05a71494aa","fromNode":"paper_cgp","fromSide":"top","toNode":"found_contact","toSide":"bottom","label":"接触建模"},
		{"id":"51de109d578c7668","fromNode":"paper_cgp","fromSide":"top","toNode":"found_signal","toSide":"bottom","label":"触觉"},
		{"id":"ee108ae1f442208d","fromNode":"paper_mcc","fromSide":"top","toNode":"found_control","toSide":"bottom","label":"柔顺控制"},
		{"id":"b2d81c7cacbb58fc","fromNode":"paper_mcc","fromSide":"top","toNode":"found_dynamics","toSide":"bottom","label":"执行器"},
		{"id":"e7a79c8c08b299b2","fromNode":"paper_dexhil","fromSide":"top","toNode":"found_embodied","toSide":"bottom","label":"VLA HiL"},
		{"id":"91c1bbec3331f31f","fromNode":"paper_dexhil","fromSide":"top","toNode":"found_rl","toSide":"bottom","label":"DAgger"},
		{"id":"9485cb946920abf6","fromNode":"paper_tacmap","fromSide":"top","toNode":"found_signal","toSide":"bottom","label":"触觉表征"},
		{"id":"e62b38a13cab3996","fromNode":"paper_tacmap","fromSide":"top","toNode":"found_compgeo","toSide":"bottom","label":"几何"},
		{"id":"2c694498a06eda82","fromNode":"paper_dapl","fromSide":"top","toNode":"found_rl","toSide":"bottom","label":"动力学RL"},
		{"id":"53a72f64388a0b94","fromNode":"paper_dapl","fromSide":"top","toNode":"found_dynamics","toSide":"bottom","label":"世界模型"},
		{"id":"9d160c384b7d0398","fromNode":"paper_gat","fromSide":"top","toNode":"found_rl","toSide":"bottom","label":"Sim2Real"},
		{"id":"0cbcf16f2b0af23e","fromNode":"paper_stola","fromSide":"top","toNode":"found_signal","toSide":"bottom","label":"触觉语言"},
		{"id":"b091314e940a2cb2","fromNode":"paper_robotwin2","fromSide":"top","toNode":"found_embodied","toSide":"bottom","label":"数据生成"},
		{"id":"e9af8477a6dbd619","fromNode":"paper_pointworld","fromSide":"top","toNode":"found_embodied","toSide":"bottom","label":"世界模型"},
		{"id":"9d799f1f68964c27","fromNode":"paper_pointworld","fromSide":"top","toNode":"found_compgeo","toSide":"bottom","label":"3D表征"},
		{"id":"8cd1a39d135e9086","fromNode":"paper_pointworld","fromSide":"top","toNode":"found_repr","toSide":"bottom","label":"Flow表征"},
		{"id":"592ea665d1736da1","fromNode":"paper_sim2real_survey","fromSide":"top","toNode":"found_rl","toSide":"bottom","label":"综述"},
		{"id":"f00cb7d18fbc2244","fromNode":"paper_sim2real_review","fromSide":"top","toNode":"found_dynamics","toSide":"bottom","label":"执行器"},
		{"id":"04dcc991dd533407","fromNode":"paper_reducers","fromSide":"top","toNode":"found_dynamics","toSide":"bottom","label":"传动"},
		{"id":"e_bt_new_200","fromNode":"bt_representation","toNode":"paper_cgp","fromSide":"bottom","toSide":"top"},
		{"id":"e_bt_new_201","fromNode":"bt_impedance","toNode":"paper_mcc","fromSide":"bottom","toSide":"top"},
		{"id":"e_bt_new_202","fromNode":"bt_sim2real","toNode":"paper_dexhil","fromSide":"bottom","toSide":"top"},
		{"id":"e_bt_new_203","fromNode":"bt_representation","toNode":"paper_tacmap","fromSide":"bottom","toSide":"top"},
		{"id":"e_bt_new_204","fromNode":"bt_sim2real","toNode":"paper_tacmap","fromSide":"bottom","toSide":"top"},
		{"id":"e_bt_new_205","fromNode":"bt_curriculum","toNode":"paper_dapl","fromSide":"bottom","toSide":"top"},
		{"id":"e_bt_new_206","fromNode":"bt_sim2real","toNode":"paper_gat","fromSide":"bottom","toSide":"top"},
		{"id":"e_bt_new_207","fromNode":"bt_representation","toNode":"paper_pointworld","fromSide":"bottom","toSide":"top"},
		{"id":"e_bt_new_208","fromNode":"bt_sim2real","toNode":"paper_sim2real_survey","fromSide":"bottom","toSide":"top"},
		{"id":"e_bt_new_209","fromNode":"bt_sim2real","toNode":"paper_sim2real_review","fromSide":"bottom","toSide":"top"},
		{"id":"e_mech_bt5","fromNode":"proj_mech_hw","fromSide":"bottom","toNode":"bt_sim2real","toSide":"top","color":"3","label":"硬件Gap"},
		{"id":"e_mech_dyn","fromNode":"proj_mech_hw","fromSide":"bottom","toNode":"found_dynamics","toSide":"top","color":"6"},
		{"id":"e_mech_ctrl","fromNode":"proj_mech_hw","fromSide":"bottom","toNode":"found_control","toSide":"top","color":"6"},
		{"id":"e_bt5_dexndm","fromNode":"bt_sim2real","fromSide":"bottom","toNode":"paper_dexndm","toSide":"top","label":"神经动力学"},
		{"id":"e_dexndm_dyn","fromNode":"paper_dexndm","fromSide":"top","toNode":"found_dynamics","toSide":"bottom","label":"System ID"},
		{"id":"e_dexndm_rl","fromNode":"paper_dexndm","fromSide":"top","toNode":"found_rl","toSide":"bottom","label":"在线适应"},
		{"id":"e_wog_embodied","fromNode":"paper_wog","fromSide":"top","toNode":"found_embodied","toSide":"bottom","label":"VLA世界模型"},
		{"id":"e_wog_repr","fromNode":"paper_wog","fromSide":"top","toNode":"found_repr","toSide":"bottom","label":"条件空间"},
		{"id":"e_bt7_wog","fromNode":"bt_representation","fromSide":"bottom","toNode":"paper_wog","toSide":"top","color":"4","label":"条件表征"},
		{"id":"e_comet_repr","fromNode":"paper_comet","fromSide":"top","toNode":"found_repr","toSide":"bottom","label":"CVAE"},
		{"id":"e_comet_stoch","fromNode":"paper_comet","fromSide":"top","toNode":"found_stochastic","toSide":"bottom","label":"GMM"},
		{"id":"e_bt4_phygile","fromNode":"bt_curriculum","fromSide":"bottom","toNode":"paper_phygile","toSide":"top","color":"5","label":"课程MoE"},
		{"id":"e_bt7_phygile","fromNode":"bt_representation","fromSide":"bottom","toNode":"paper_phygile","toSide":"top","label":"TP-MoE"},
		{"id":"e_phygile_rl","fromNode":"paper_phygile","fromSide":"top","toNode":"found_rl","toSide":"bottom","label":"PPO+课程"},
		{"id":"e_phygile_repr","fromNode":"paper_phygile","fromSide":"top","toNode":"found_repr","toSide":"bottom","label":"扩散生成"},
		{"id":"e_phygile_dyn","fromNode":"paper_phygile","fromSide":"top","toNode":"found_dynamics","toSide":"bottom","label":"262D机器人"},
		{"id":"e_bt5_rlt","fromNode":"bt_sim2real","fromSide":"bottom","toNode":"paper_rlt","toSide":"top","color":"5","label":"在线适应"},
		{"id":"e_rlt_rl","fromNode":"paper_rlt","fromSide":"top","toNode":"found_rl","toSide":"bottom","label":"在线RL"},
		{"id":"e_rlt_embodied","fromNode":"paper_rlt","fromSide":"top","toNode":"found_embodied","toSide":"bottom","label":"VLA后训练"},
		{"id":"e_rlt_repr","fromNode":"paper_rlt","fromSide":"top","toNode":"found_repr","toSide":"bottom","label":"信息瓶颈"},
		{"id":"e_rlt_wmpo","fromNode":"paper_rlt","fromSide":"left","toNode":"paper_wmpo","toSide":"right","label":"互补: 轻量vs全模型"},
		{"id":"e_bt7_act","fromNode":"bt_representation","fromSide":"bottom","toNode":"paper_act","toSide":"top","color":"4","label":"双臂操作"},
		{"id":"e_act_repr","fromNode":"paper_act","fromSide":"top","toNode":"found_repr","toSide":"bottom","color":"6","label":"CVAE"},
		{"id":"e_act_embodied","fromNode":"paper_act","fromSide":"top","toNode":"found_embodied","toSide":"bottom","color":"6","label":"模仿学习"},
		{"id":"e_bt5_recap","fromNode":"bt_sim2real","fromSide":"bottom","toNode":"paper_recap","toSide":"top","color":"4","label":"VLA后训练"},
		{"id":"e_recap_rl","fromNode":"paper_recap","fromSide":"top","toNode":"found_rl","toSide":"bottom","color":"6","label":"经验RL"},
		{"id":"e_recap_embodied","fromNode":"paper_recap","fromSide":"top","toNode":"found_embodied","toSide":"bottom","color":"6","label":"VLA"},
		{"id":"e_recap_wmpo","fromNode":"paper_recap","fromSide":"left","toNode":"paper_wmpo","toSide":"right","color":"3","label":"互补: 经验vs世界模型"},
		{"id":"e_recap_rl100","fromNode":"paper_recap","fromSide":"left","toNode":"paper_rl100","toSide":"right","color":"3","label":"互补: VLA vs 扩散RL"},
		{"id":"e_bt2_unified","fromNode":"bt_sparse","fromSide":"bottom","toNode":"paper_unified","toSide":"top","color":"4","label":"RL统一理论"},
		{"id":"e_unified_rl","fromNode":"paper_unified","fromSide":"top","toNode":"found_rl","toSide":"bottom","color":"6","label":"分类框架"},
		{"id":"e_unified_optim","fromNode":"paper_unified","fromSide":"top","toNode":"found_optim","toSide":"bottom","color":"6","label":"优化理论"},
		{"id":"e_proj_meeting_pipeline","fromNode":"proj_dnpm_insight","fromSide":"right","toNode":"proj_meeting_pipeline","toSide":"left","color":"1","label":"会议综合"},
		{"id":"e_pipeline_bt3","fromNode":"proj_meeting_pipeline","fromSide":"top","toNode":"bt_impedance","toSide":"bottom","color":"2","label":"分阶段落地"}
	]
}