The authors diagnose a failure mode in autoregressive robot video world models that use full history decoding (Full Context): as rollout length increases, attention on recent visual states declines, leading to degraded predictions.
Context scheduling improves long-horizon robot video prediction
ReCAP method preserves initial scene and recent frames while discarding stale history, boosting fidelity and efficiency without retraining.
Chinese Tech
Hao Wu · Tianyu Shao · Shengju Qian · Yizhou Zhao · Fan Zhang · Penghao Zhao · +7 more
The Hong Kong University of Science and Technology (Guangzhou) · North China Electric Power University · LIGHTSPEED · Carnegie Mellon University · The Chinese University of Hong Kong
Research Digest··3 min read
Thread:Robot World Models
Hao Wu and colleagues from multiple institutions introduce ReCAP, a training-free context-scheduling method for autoregressive robot video world models.
Why this paper
From Tencent and 8 others · Part of Robot World Models, now 27 papers
In one line
Reorganizing inference context by preserving initial anchor and recent blocks improves long-horizon robot video world model prediction and efficiency.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (2 noted)
- ✓Reports numbers on named benchmarks (3 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§