ACID: Action Consistency via Inverse Dynamics for Planning with World Models

arXiv 2026 35.6 method

TLDR

ACID improves decision-time planning with world models by enforcing cycle action consistency via inverse dynamics, enhancing efficiency across diverse tasks.

Reasoning

The paper introduces a novel consistency constraint for world model planning, validated across multiple models and tasks, showing clear improvements. However, it is limited to decision-time planning and may rely on accurate inverse dynamics, and the abstract does not specify real-world deployment.

Read-first score

Read-first score 35.6, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 37.

Recency 6%

100

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

Topical relevance 29%

52.9

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 18%

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

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Citation impact 18%

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

Citation velocity 12%

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

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 177.

Keyword Scores

world model

world dynamics prediction

model-based reinforcement learning world model

world simulator

generative world model

interactive world model

video world model

Deep Analysis

Innovations

Cycle action consistency: enforcing that the action inferred backward from a predicted transition by an inverse dynamics model recovers the conditioned action
Integrating per-step inverse dynamics residuals into the planning cost with a scale-invariant adaptive weight

Methodology

ACID adds a cycle consistency term to decision-time planning: for each predicted transition, an inverse dynamics model infers the action, and the residual between inferred and conditioned action is penalized. This per-step residual is combined with the standard goal-distance cost using a scale-invariant adaptive weight, and evaluated across four action-conditioned world models on six embodied control tasks.

Key Results

ACID consistently improves planning performance over baselines and matches baseline accuracy with substantially less planning compute across rigid/deformable manipulation, articulated control, and visual navigation.

Resources

Paper