AI TrackerFrontier desk · AEST
Back to news
normalaitech

Long-WAM: Scaling the Context of World-Action Models

arXiv cs.AI · 8 Oct 2026, 03:58 AEST

Long-WAM is a model-system framework designed to enhance real-time robot control by scaling the context of causal world-action models. It demonstrates significant improvements in task success rates by effectively utilizing longer visual histories during action prediction.

Key facts

Read aloud

Long-WAM is a model-system framework designed to enhance real-time robot control by scaling the context of causal world-action models. It demonstrates significant improvements in task success rates by effectively utilizing longer visual histories during action prediction. Real-time robot control demands enough visual history to infer motion and task progress, but processing that history can delay action. Our central finding is that access to history is not the same as using it: longer histories pay off far more when the video foundation is pretrained autoregressively AR. On RoboCasa GR-one, increasing context from zero point zero to nineteen point two seconds raises success from sixty three point three percent to seventy eight point seven percent, whereas a bidirectionally pretrained initialization shows no net gain. Streaming observation encoding, asynchronous execution, and hardware-specific acceleration enable deployment on RTX five thousand ninety, DGX Spark, and Jetson AGX Thor without dropping future prediction. Real-time deployment on Unitree G1 and YAM supports dynamic and long-horizon manipulation, including ninety five percent success on dynamic cup stacking.

Sources in this cluster