Video World Models · 2026-10-14
RAR: Rate-Adaptive Rollout for Time-Consistent Interactive World Models
Yaoyang Liu, Dongyu Yan, Yihua Du, Shengju Qian, Zeke Xie, Xin Wang, Ying-Cong Chen, Long Chen
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
RAR (Rate-Adaptive Rollout) addresses temporal-rate coupling in interactive world models by enabling controllable rollout temporal rates. It uses Stride-RoPE to align latent frames sampled at different strides on a shared physical-time coordinate, supports multi-stride training via multi-stride video sampling, and allows online rate switching by varying stride during rollout without discarding history. Experiments report temporally consistent multi-rate rollout and online rate switching across Doom, CS2, and FPV.
Why it matters
Interactive world models must advance in real time under varying runtime throughput, but fixed-rate training/rollout can cause temporal mismatch (rate–throughput coupling) and forces a trade-off between temporal granularity and physical horizon. RAR targets both issues by learning and controlling dynamics across multiple temporal rates using a shared physical-time representation, aiming for time-consistent user interaction and longer physical horizons.
Method
- Introduces Rate-Adaptive Rollout (RAR) that parameterizes temporal rate by a stride and supports multi-rate training and inference.
- Proposes Stride-RoPE: reparameterizes RoPE in physical-time units so equal physical-time intervals induce equal relative rotary phases across sampling rates.
- Uses multi-stride video sampling for joint learning across fine local dynamics and long-horizon evolution; at inference, changes stride online without discarding rollout history (KV cache rate-aware via accumulated stride).
Limitation
The authors report that extrapolation beyond the training range lies outside the scope of their claims; SR at out-of-range rates is provided only to characterize the boundary of the method.
Abstract (from arXiv)
Video-based interactive world models aim to enable real-time, user-controllable environments. Yet most existing systems are trained and rolled out at a fixed temporal rate, limiting both runtime-rate adaptation and physical-horizon coverage -- a limitation we term temporal-rate coupling. Specifically, when runtime generation throughput differs from the learned temporal rate, the generated world evolves unintentionally more slowly or quickly. Moreover, using a single temporal rate imposes an inherent trade-off between temporal granularity and physical horizon: higher rates capture finer motion but cover a shorter duration under a fixed context window. To address both limitations, we introduce Rate-Adaptive Rollout (RAR), a framework that enables interactive world models to roll out at controllable temporal rates. At its core, Stride-RoPE aligns latent frames sampled at different rates on a shared physical-time coordinate, enabling consistent multi-rate rollout, unseen-rate generalization, and online rate switching without discarding history. Meanwhile, RAR uses multi-stride video sampling to enable joint learning of fine-grained dynamics and long-horizon evolution. Across action- and camera-controlled environments, RAR achieves temporally consistent multi-rate rollout and online rate switching while preserving high-rate generation quality and extending physical horizons under matched training budgets. RAR also transfers to pretrained world models while preserving perceptual quality.
Related papers
- Time-JEPA: Grounding Video World Models with Physical Time Series
- LongForcing: Scaling Temporal Manifold Matching for Long-Horizon World Rollout
- LongTake: Learning to Sustain Dynamics in Long-Horizon Video Generation
- Oneira: From Open-Ended Generation to Open-World Interaction in Video World Models
- StateCraft: Augmenting Video World Models with Abstract-State Reasoning
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-12 · Indexed by Pub-AI: 2026-10-14 UTC
RAR: Rate-Adaptive Rollout for Time-Consistent Interactive World Models
Preprint posted: 2026-10-12 · Original record ↗
- arxiv
- 2610.16302
Explicit source versions (1)
- v1 · 2026-10-12
- Preprint available
Source date: 2026-10-12 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
Video-based interactive world models aim to enable real-time, user-controllable environments. Yet most existing systems are trained and rolled out at a fixed temporal rate, limiting both runtime-rate adaptation and physical-horizon coverage -- a limitation we term temporal-rate coupling. Specifically, when runtime generation throughput differs from the learned temporal rate, the generated world evolves unintentionally more slowly or quickly. Moreover, using a single temporal rate imposes an inherent trade-off between temporal granularity and physical horizon: higher rates capture finer motion but cover a shorter duration under a fixed context window. To address both limitations, we introduce Rate-Adaptive Rollout (RAR), a framework that enables interactive world models to roll out at controllable temporal rates. At its core, Stride-RoPE aligns latent frames sampled at different rates on a shared physical-time coordinate, enabling consistent multi-rate rollout, unseen-rate generalization, and online rate switching without discarding history. Meanwhile, RAR uses multi-stride video sampling to enable joint learning of fine-grained dynamics and long-horizon evolution. Across action- and camera-controlled environments, RAR achieves temporally consistent multi-rate rollout and online rate switching while preserving high-rate generation quality and extending physical horizons under matched training budgets. RAR also transfers to pretrained world models while preserving perceptual quality.