Act2Goal: From World Model To General Goal-conditioned Policy

arXiv 25.12 2025 57.2 method, application

TLDR

Act2Goal integrates a goal-conditioned visual world model with multi-scale temporal control for long-horizon robotic manipulation, achieving zero-shot generalization and online adaptation.

Reasoning

The paper presents a novel integration of a world model for generating intermediate visual states and a multi-scale temporal hashing mechanism for robust execution. Strengths include real-robot experiments and zero-shot generalization; weaknesses are that the abstract is cut off, limiting full assessment of results and limitations.

Read-first score

Read-first score 57.2, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 47.

Recency 8%

86.7

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

Topical relevance 42%

67.1

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 25%

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

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

Frontier

Rank sensitivity

Stability: volatile; rank range: 375.

Keyword Scores

world model

generative world model

world dynamics prediction

video world model

model-based reinforcement learning world model

world simulator

interactive world model

Deep Analysis

Innovations

Integration of a goal-conditioned visual world model with multi-scale temporal control for long-horizon manipulation
Multi-Scale Temporal Hashing (MSTH) that decomposes imagined trajectories into dense proximal frames for fine-grained closed-loop control and sparse distal frames for global task consistency
End-to-end cross-attention mechanism coupling visual representations with motor control for coherent long-horizon behavior
Reward-free online adaptation via hindsight goal relabeling with LoRA-based finetuning, enabling rapid autonomous improvement without external supervision

Methodology

Act2Goal uses a goal-conditioned visual world model to generate a plausible sequence of intermediate visual states from the current observation to a target visual goal. The Multi-Scale Temporal Hashing (MSTH) decomposes this imagined trajectory into dense proximal frames for fine-grained closed-loop control and sparse distal frames for global task consistency. The policy then couples these representations with motor control through end-to-end cross-attention, enabling coherent long-horizon behavior while remaining reactive to local disturbances. The method also incorporates reward-free online adaptation through hindsight goal relabeling with LoRA-based finetuning.

Key Results

In real-robot experiments, Act2Goal improves success rates from 30% to 90% on challenging out-of-distribution tasks within minutes of autonomous interaction, demonstrating strong zero-shot generalization to novel objects, spatial layouts, and environments.

Resources

Paper