The authors propose Long-WAM, a model-system framework for scaling the context of world-action models under real-time control constraints.
Autoregressive video pretraining unlocks the value of long context for real-time robot control
Long-WAM scales visual history up to 19.2 seconds for action prediction, achieving high success on dynamic tasks.
Big Tech
Wei Huang · Bohan Zhang · Chenzhi Liu · Isabella Liu · Shuai Yang · Weian Mao · +10 more
NVIDIA · MIT · HKU · UCSD
Research Digest··3 min read
The authors introduce Long-WAM, a framework that scales the context window of causal world-action models for real-time robot control.
Why this paper
From NVIDIA and 3 others
In one line
Longer visual history improves world-action model control only with autoregressive video pretraining.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ✓Compute or model size stated (hardware RTX 5090)
- ·No stated limitations found
- ✓Reports numbers on named benchmarks (9 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§