Maker's videoGalbot · 19 July 2026
WAM-TTT: Teaching Robots to Adapt to New Environments by Watching Humans | GALBOT AstraBrain
Against the record
Galbot G1, from Galbot (Galaxy General Robot), is at stage 3 of 5 on our ladder: pilot. Working with a named customer in a real setting. The latest milestone on its record is from Dec 2025: Galbot said G1 helps in patient rooms, pharmacies and with directions at Xuanwu Hospital.
This video comes from the maker's own channel, and makers show their robots at their best. The stage above comes from dated public records, not from the video.
From the video's description
Galbot introduces WAM-TTT (World-Action Model Test-Time Training), a new test-time training framework within AstraBrain that enables scalable deployment of humanoid robots. Today, deploying robots to new environments often requires costly robot teleoperation data collection, extensive manual annotations, and repeated fine-tuning. WAM-TTT introduces a new paradigm: robots can adapt to new scenarios by learning from human demonstration videos, while the pretrained foundation model remains frozen and only lightweight fast weights are adapted. Through real-world robot evaluations, WAM-TTT demonstrates strong generalization across diverse environments, object variations, and lighting conditions, significantly outperforming in-context learning (ICL). Key breakthroughs: • The world’s first test-time training paradigm for embodied AI enabling deployment adaptation without robot action data • …
Written by Galbot on YouTube.
More videos of Galbot G1
Robot Tennis Stuns the Crowd in Seoul 🎾🤖 | GALBOT’s Korean Debut
Tennis Pro vs. a Humanoid Robot 🎾🤖 | China Open
From AlphaGo to AstraTennis: The World’s First Autonomous Humanoid Robot Tennis Match | GALBOT
Meet GALBOT ET1.
World’s First General-Purpose Cerebellum GPT Foundation Model for Humanoid Robots: AstraBrain-WBC0.5
All Data. Any Quality. Fully Utilized by one Latent World Action Model