NVIDIA releases Long-WAM, a world-action model scaling visual context for real-time robot control
_akhaliq · x · 2026-10-08
NVIDIA has released Long-WAM, a world-action model for real-time robot control that scales visual context to leverage longer observation histories.
A key empirical finding: access to history is not the same as using it — longer histories pay off far more when the video foundation model is pretrained autoregressively. The result offers direct guidance on pretraining strategies for robot learning.
Related event: NVIDIA Open-Sources Long-WAM for Real-Time Robot Control(2 posts)→
More from Embodied
- multimodalart gets hands on a microduck prototype hardware — multimodalart · 2026-10-09
- Interface, a $99 Handheld Controller for AI Agents, Opens Waitlist Ahead of January Ship — cneuralnetwork · 2026-10-09
- Apple announces surprise Oct 13 'Welcome home' event: smart home hub, Apple TV, LG devices — tomwarren · 2026-10-09
- Tesla Robotaxi team chased storms 24/7 to validate all-weather autonomous driving — yunta_tsai · 2026-10-09
- Agent Buddy: an ESP32 desk pet that shows what your coding agents are up to — DanWahlin · 2026-10-09
- Viture Beast on iPhone 18: reviewer says audio-only beats glasses for ambient compute — curious_vii · 2026-10-08