DeltaWAM: Delta World Action Models for Bimanual Manipulation

arXiv 2026 36.7 method, application

TLDR

DeltaWAM improves bimanual manipulation by predicting visual deltas and actions with dense-anchor, sparse-delta, and action streams plus Streaming Delta Memory, boosting efficiency and success.

Reasoning

The paper presents a novel world-action model that reduces redundant visual prediction and inference cost while improving task success in simulation and real-world evaluations. Strengths include clear efficiency gains and public code/website; limitations are that the abstract omits detailed real-world setup and per-architecture analysis.

Read-first score

Read-first score 36.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 46.

Recency 5%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Topical relevance 25%

65.7

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Methodology quality 15%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=evaluation

Reproducibility 15%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=code,github

Graph prestige 15%

5.7

PageRank-style prestige over the local citation graph built from available OpenAlex references. local_pagerank=0.0016

Citation impact 15%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 10%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 388.

Keyword Scores

world model

world dynamics prediction

video world model

generative world model

interactive world model

world simulator

model-based reinforcement learning world model

Deep Analysis

Innovations

DeltaWAM jointly predicts visual deltas and actions using dense-anchor, sparse-delta, and action streams, avoiding dense future-frame prediction and decoupling action-conditioned dynamics from nuisance appearance variations.
Introduces three DeltaWAM architectures that differ in representation and computation sharing.
Streaming Delta Memory (SDM) updates cached anchor context with compact observed deltas, reducing heavy video-expert processing during inference.

Methodology

DeltaWAM builds on world-action models that transfer visual and motion priors from pretrained video generators to robot control. It uses dense-anchor, sparse-delta, and action streams to jointly predict visual deltas and actions, with three architectures varying representation and computation sharing, plus Streaming Delta Memory to update cached anchor context with compact observed deltas. Evaluation on RoboTwin and real-world bimanual manipulation compares against Fast-WAM using average success, normalized progress, training FLOPs, and inference latency/FLOPs.

Key Results

DeltaWAM with SDM improves average success over Fast-WAM from 81.3% to 85.4% in the clean setting and from 75.8% to 83.9% under visual randomization. The three architectures reduce training FLOPs by 17.78-23.77%, while SDM reduces one-step inference latency and FLOPs by 36.57% and 31.55%, respectively; real-world evaluations show the highest overall success rate and normalized progress among evaluated policies.

Resources

Paper