Physical AI 前沿雷达 · 每周汇总 · 2026-W38
2026-09-14 → 2026-09-20 共发布 15 条，按主线分组。标题链接到原始来源，日期链接到当天的雷达。
- 证据标注：O = 一手官方 · R = 论文/预印本 · M = 媒体或二手
[O]3 ·[R]7 ·[M]5
基座模型（VLA / WAM） (3)
[R]GeoAAC: Geometry-Based Adaptive Action Chunking from Denoising Trajectories in VLA Policies — arXiv 2609.20776v1 ｜ 2026-09-19[R]Agile-WAM: An Agile Tactile World Action Model for Contact-Rich Robot Control — arXiv 2609.20761v1 ｜11.9 ms｜ 2026-09-19[O]Fine-Tuning π0 (Pi-Zero) for robotic manipulation on Amazon SageMaker HyperPod EKS — AWS Physical AI Blog ｜ 2026-09-20
数据引擎（采集 / 清洗 / scaling） (1)
[M]Robots Are Learning to Feel — IEEE Spectrum Robotics ｜100 hours·12 tasks·3,000 hours｜ 2026-09-19
训练与自我改进（RL / 后训练） (2)
[R]Towards High-DoF Dexterous Manipulation through VLA Post-Training — arXiv 2609.19666v1 ｜ 2026-09-19[O]Cute Critters Come to the Cloud: ‘Aniimo’ Launches on GeForce NOW — NVIDIA Blog ｜ 2026-09-20
仿真与评测（sim2real / 闭环） (3)
[R]FIVE-VLA: Fast and EffectIVE Autonomous Driving with Recurrent Action Memory — arXiv 2609.18623v1 ｜641M parameters·10%·10.2%｜ 2026-09-19[M]Video Friday: Two Birotors Make a Quadrotor — IEEE Spectrum Robotics ｜33%·2x｜ 2026-09-20[M]Video Friday: Digit Redecorates — IEEE Spectrum Robotics ｜ 2026-09-20
边缘与实时性（端侧推理 / 控制频率） (3)
[R]rMuscle: Robotic Muscle Memory for Efficient Vision-Language-Action Model Inference — arXiv 2609.19104v1 ｜ 2026-09-19[R]PASSAGE: Scaling Scene-Aligned Motion Learning for Perceptive Humanoid Traversal in Cluttered Environments — arXiv 2609.18732v1 ｜50 Hz·48.1%·68.9%｜ 2026-09-19[O]NVIDIA Vera Rubin NVL72 Delivers Leading Performance in MLPerf Inference v6.1 Debut — NVIDIA Blog ｜ 2026-09-20
系统与编排（长程 / 多策略） (1)
[R]StageGuard: Learning Stage Transitions for Long-Horizon Robot Tasks via Agentic Distillation — arXiv 2609.20791v1 ｜ 2026-09-19
本体与供应链（硬件 / 成本） (2)
[M]Digit 5 May Be the First Humanoid Robot Worker That’s Truly Safe — IEEE Spectrum Robotics ｜20 hours｜ 2026-09-20[M]Video Friday: Meet Microduck — IEEE Spectrum Robotics ｜ 2026-09-20
本仓库为技术观察，不构成投资建议、产品承诺或安全认证结论。论文结论以作者自报为准。