SCAR: Self-Supervised Continuous Action Representation Learning

arXiv 2026 51.4 method

TLDR

SCAR learns unified action representations across embodiments via joint inverse-forward dynamics, improving world model generalization and cross-embodiment adaptation.

Reasoning

The paper introduces a novel self-supervised framework with regularization and adversarial invariance to learn transferable action representations, which is a strength. However, evaluation is limited to Procgen and Robotwin datasets, and reliance on a pretrained generative backbone may limit applicability.

Read-first score

Read-first score 51.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 29.

Recency 6%

100

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

Citation impact 18%

80.8

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.80810256

Methodology quality 18%

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

Topical relevance 29%

41.4

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

Reproducibility 18%

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

Citation velocity 12%

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

Field roles

FoundationFrontierBridgeMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 280.

Keyword Scores

world model

world dynamics prediction

generative world model

model-based reinforcement learning world model

interactive world model

video world model

world simulator

Deep Analysis

Innovations

Proposes SCAR, a joint inverse-forward dynamics framework for learning unified action representations across embodiments from visual transitions.
Argues that action is a distinct representational factor decoupling controllable change from embodiment-specific actuation, not merely an auxiliary conditioning signal.
Introduces regularization of latent action posterior toward a standard Gaussian prior to limit arbitrary visual encoding and adversarial invariance to suppress embodiment- and environment-specific nuisance factors.

Methodology

SCAR uses a pretrained generative backbone with an inverse dynamics model (IDM) to infer latent actions from latent observation pairs and a forward dynamics model (FDM) to predict future dynamics conditioned on those latent actions. The latent action space is regularized with a standard Gaussian prior and adversarial invariance to make representations transferable across embodiments.

Key Results

Experiments on Procgen and Robotwin datasets show that the learned unified latent action representation serves as a stronger conditioning interface for world modeling than embodiment-specific raw actions, yielding improved cross-embodiment low-data adaptation and cross-task transfer.

Resources

Paper