ResearchNovember 18, 2026
22
🌶️
Robot policy survives a bumped camera
WACO makes VLA robots pose-robust: +36.7 success points on real UR5e.
#Robotics#VLA#Few-Shot#Computer Vision#arXiv

🔥 What happened
Researchers from Shandong and partner institutes unveil WACO: a VLA model that self-calibrates from a single RGB-D frame when the camera or robot base has shifted. No ground-truth extrinsics, no extra data. They also ship LIBERO-PoseShift, the first benchmark for graded camera, base, and compound shifts.
💡 Why it matters
The standard few-shot recipe silently breaks when a tripod gets bumped or a base re-docks a few centimeters off. WACO lifts shifted success by 13.2 points in simulation and 36.7 points on a real UR5e, closing 90% of the gap to an oracle that knows the true pose.
⚡ Our take
This is the unglamorous glue that finally gets robot demos out of the lab. Anyone still assuming perfectly aligned cameras is building for a world that doesn't exist.
The title, summary and analysis of this item were produced automatically by an AI system and have not been editorially reviewed. They may contain errors, bias or omissions — when in doubt, read the linked original source.
Deep Dives & Similar Intelligence

Caching barely pays off for robot policies
ResearchNovember 15, 2026

NVIDIA's BirdDreams predicts traffic from one image
ResearchNovember 16, 2026

LLM Agents Self-Improve Without Instructions
ResearchNovember 17, 2026

Backdoor Attack Survives Continual Learning
Ethics & SecurityNovember 14, 2026

MoE-Context: AI Agents Learn Selective Reading
ResearchNovember 18, 2026