Active Intelligence in Video Avatars via Closed-loop World Modeling

arXiv 25.12 2025 61 method, benchmark, system

TLDR

Introduces ORCA, a closed-loop world modeling framework for video avatars to achieve autonomous goal-directed behavior via hierarchical reasoning and belief updating.

Reasoning

The paper presents a novel framework (ORCA) and benchmark (L-IVA) that address the lack of agency in video avatars by integrating internal world models with a closed-loop Observe-Think-Act-Reflect cycle. Strengths include a clear problem formulation (POMDP) and demonstrated improvements over baselines. Weaknesses are the lack of explicit real-world evaluation details and reliance on generative environments, which may limit generalizability.

Read-first score

Read-first score 61, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 45.

Recency 8%

86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Methodology quality 25%

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

Topical relevance 42%

64.3

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 25%

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 200.

Keyword Scores

world model

video world model

interactive world model

world dynamics prediction

generative world model

model-based reinforcement learning world model

world simulator

Deep Analysis

Innovations

Closed-loop OTAR cycle (Observe-Think-Act-Reflect) for robust state tracking under generative uncertainty
Hierarchical dual-system architecture with System 2 (strategic reasoning) and System 1 (action caption translation)
Formulation of avatar control as a POMDP with continuous belief updating and outcome verification

Methodology

ORCA is a framework that embodies an Internal World Model (IWM) through a closed-loop OTAR cycle and a hierarchical dual-system architecture. It formulates avatar control as a POMDP, implementing continuous belief updating with outcome verification to enable autonomous multi-step task completion in open-domain scenarios. The framework is evaluated against open-loop and non-reflective baselines on the L-IVA benchmark.

Key Results

ORCA significantly outperforms open-loop and non-reflective baselines in task success rate and behavioral coherence.

Resources

Paper Project