Galaxy Universal Releases Embodied TTT Framework
量子位 · wechat · 2026-07-16
Galaxy Universal has released **WAM-TTT**, the world's first test-time post-training framework for embodied AI. It migrates the common Test-Time Training concept from large models into robot control, allowing robots to "learn and apply on the fly" during deployment. The core mechanism involves first understanding the current task using human demonstration videos, then writing the task information into a lightweight `fast-weight memory`. The main model weights remain frozen; only the memory module is updated. Officially, this approach significantly reduces deployment costs without relying on massive robot trajectories or manual action annotations, while mitigating cross-scene generalization and catastrophic forgetting issues. The article also shares comparative results: when combining human videos with robot trajectories, the task success rate approaches that of training purely on robot trajectories. Compared to ICL, LoRA, and direct mixed-training methods, WAM-TTT demonstrates greater stability across multiple tasks. The author views this as a crucial step into the "post-training era" of embodied AI.
Related event: Galactic General Releases WAM-TTT for Embodied AI(2 posts)→
More from Embodied
- Snake-like robot moves on fully passive wheels and winding motion — ___Mufasaa · 2026-07-21
- Creator buys a Reachy robot and asks what to build first — dee_hw · 2026-07-21
- MW team shows Gen1 of MW-bot, a semi-humanoid home robot built for pantry storage and ceiling rails — CyberRobooo · 2026-07-21
- Kunlun Spins Out Riemann Dynamics, Unveils World Action Model for Robotics — 0xAllen_ · 2026-07-21
- Gritt exits stealth with $34M to automate construction sites, starting with solar plants — TechCrunch AI · 2026-07-21
- BrainCo demos near-real-time bionics without implants and claims 85% lower prosthetic cost — TrueOrange9944 · 2026-07-21