Robotics researchers mock low-data VLAs: 30 demos is just visual servoing
KyleMorgenstein · x · 2026-10-12
Robotics researcher @sincethestudy quote-tweets a demo built as a UMI-based ACT policy trained on only 30 demonstrations, arguing that VLAs trained on low data are effectively just visual servoing. The thread captures a growing skepticism in the robotics community about how much of recent low-data VLA demos reflects genuine generalization versus classic hand-tuned visual tracking.
More from Embodied
- Shenzhen robot map: 36 companies, most humanoid robot firms of any Chinese city — FinanceYF5 · 2026-10-12
- Hangzhou robot map: 25 companies led by Unitree, fastest productization in China — FinanceYF5 · 2026-10-12
- Shanghai robot map: 25 companies, industrial heavyweights FANUC, ABB, KUKA lead — FinanceYF5 · 2026-10-12
- Silicon Valley hosts 67 robot companies vs Shenzhen's 36: a 5-map comparison — FinanceYF5 · 2026-10-12
- EmbodiedCLBench: First Benchmark for Continual Learning in Embodied Agents — PKU-VaLuE-Lab · 2026-10-12
- Hive Robotics trains Unitree G1 to restock supermarket shelves, autonomous by 2027 — Distinct-Question-16 · 2026-10-12