⌁ Awesome World Model Hub Papers · Datasets · Projects
中文 GitHub
← Back to papers

Planning from Pixels using Inverse Dynamics Models

arXiv 2020 43.8 method

TLDR

Learning task-conditioned inverse dynamics models to predict action sequences for planning from pixels in model-based RL.

Reasoning

The paper proposes a novel approach to learn latent world models by predicting future actions conditioned on task completion, which focuses modeling on task-relevant dynamics. Strengths include a clear methodology and empirical evaluation on visual goal completion tasks, but the abstract lacks details on limitations and comparisons to other model-based methods.

Read-first score

Read-first score 43.8, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 34.

Methodology quality 25%
50

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

Topical relevance 42%
48.6

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

Recency 8%
42.4

Uses a gentle age decay so recent papers surface without erasing older foundations. 2020

Reproducibility 25%
30

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

Field roles

Candidate

Rank sensitivity

Stability: volatile; rank range: 220.

Keyword Scores

world model
9
model-based reinforcement learning world model
8
world dynamics prediction
6
interactive world model
5
video world model
3
generative world model
2
world simulator
1

Deep Analysis

Innovations

  • Learning latent world models by predicting future action sequences conditioned on task completion (inverse dynamics)
  • Task-conditioned models that adaptively focus on task-relevant dynamics and serve as a planning heuristic for sparse rewards

Methodology

The method learns latent world models by predicting sequences of future actions conditioned on task completion, using inverse dynamics. These models focus on task-relevant dynamics and are used as a planning heuristic for sparse reward tasks. Evaluation is on visual goal completion tasks from pixels, comparing to model-free approaches.

Key Results

The method shows a substantial increase in performance on visual goal completion tasks compared to prior model-free approaches.

Resources

Paper

Tags