Driving · 2026-11-11
Elastic-WAM: Adaptive Spatiotemporal Imagination for Efficient Robot Control
Haitian Li, Tianfu Li, Xuxi Zhao, Zizheng Liu, Yangcheng Yu, Yiwen Hou, Zeyu Xiao, Mutian Xu, Ziqian Yang, Haonan Chen, Lin Shao
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
Elastic-WAM adapts the amount of future-world refinement during robot control by learning a unified spatiotemporal “refinement demand” field that decides when to stop, how far to advance denoising, and which future video regions to refresh. It uses a frozen WAM backbone with action-grounded supervision and recurrent sparse world refinement at inference.
Why it matters
Robot control with world-action models can be slow because iterative future video denoising has substantial inference overhead. Elastic-WAM addresses the limitation of fixed refinement budgets by learning state-dependent refinement decisions over both temporal denoising progression and spatial future tokens, aiming to improve task success while reducing inference latency.
Method
- Proposes a unified spatiotemporal refinement demand field that governs WHEN to stop world refinement, HOW FAR to advance denoising, and WHERE to refresh future video regions.
- Parameterizes the field with a shared temporal basis and region-specific spatial amplitudes; learns it via three-stage action-grounded supervised learning while keeping the pretrained WAM frozen.
- At inference, starts with a dense joint denoising step, then uses the learned field for recurrent sparse world refinement and continues action denoising after world refinement stops using cached representations.
Abstract (from arXiv)
World-Action Models improve robot control by predicting future visual dynamics, but iterative video denoising incurs substantial inference overhead. Recent approaches reduce this cost through partial future denoising, yet typically rely on fixed refinement budgets. Our empirical observations reveal that preferred refinement depths vary across tasks and states, while different future regions exhibit varying relevance to action generation. Motivated by these findings, we propose Elastic-WAM, which adaptively allocates world refinement computation across temporal denoising time and spatial future space. A unified spatiotemporal refinement demand field governs when to stop world refinement, how far to advance each denoising step, and which regions to refresh. We parameterize this field with a shared temporal basis and spatial amplitudes, and learn it through a three-stage action-grounded supervised procedure. At inference, the learned field guides recurrent sparse world refinement to improve task success while reducing inference overhead. Evaluations across diverse simulation benchmarks and real-world long-horizon manipulation demonstrate state-of-the-art control performance with reduced inference latency.
Related papers
- Sparse-WAM: Accelerating World Action Models via Action-Guided Sparse Imagination
- {\Delta}WAM: Distilling Action Tangent Fields into World Action Models
- MARS: Systematic Multi-Level Acceleration for World Action Models
- W2A: Distilling World Priors into Action Models for Real-Time Robotic Control
- HiLo-WAM: Decoupling Long-Horizon Task Evolution and Local Dynamics in World Action Models
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-11-10 · Indexed by Pub-AI: 2026-11-11 UTC
Elastic-WAM: Adaptive Spatiotemporal Imagination for Efficient Robot Control
Preprint posted: 2026-11-10 · Original record ↗
- arxiv
- 2611.14257
Explicit source versions (1)
- v1 · 2026-11-10
- Preprint available
Source date: 2026-11-10 · First detected: 2026-11-11 UTC · Historical baseline
Source evidence (2)
- Official evidence ↗ · RSS guid=oai:arXiv.org:2611.14257v1; arxiv:announce_type=new; item pubDate (official metadata)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
World-Action Models improve robot control by predicting future visual dynamics, but iterative video denoising incurs substantial inference overhead. Recent approaches reduce this cost through partial future denoising, yet typically rely on fixed refinement budgets. Our empirical observations reveal that preferred refinement depths vary across tasks and states, while different future regions exhibit varying relevance to action generation. Motivated by these findings, we propose Elastic-WAM, which adaptively allocates world refinement computation across temporal denoising time and spatial future space. A unified spatiotemporal refinement demand field governs when to stop world refinement, how far to advance each denoising step, and which regions to refresh. We parameterize this field with a shared temporal basis and spatial amplitudes, and learn it through a three-stage action-grounded supervised procedure. At inference, the learned field guides recurrent sparse world refinement to improve task success while reducing inference overhead. Evaluations across diverse simulation benchmarks and real-world long-horizon manipulation demonstrate state-of-the-art control performance with reduced inference latency.
