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

Terra ACT-Bench: Towards Action Controllable World Models for Autonomous Driving

arXiv 24.12 2024 77.5 method, benchmark, system, application

TLDR

An open-source benchmark and baseline model for evaluating action fidelity in world models for autonomous driving.

Reasoning

The paper addresses a clear gap in evaluating action controllability of world models, providing an open-access framework and baseline. Its strength lies in reproducibility and focus on action fidelity, but it is limited to the autonomous driving domain and does not cover other aspects of world model evaluation.

Read-first score

Read-first score 77.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 52.

Reproducibility 25%
81

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

Methodology quality 25%
80

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

Recency 8%
75.1

Uses a gentle age decay so recent papers surface without erasing older foundations. 2024

Topical relevance 42%
74.3

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

Field roles

Methodology anchorReproducibility anchor

Rank sensitivity

Stability: volatile; rank range: 32.

Keyword Scores

world model
10
world dynamics prediction
9
generative world model
8
video world model
8
interactive world model
7
world simulator
6
model-based reinforcement learning world model
4

Deep Analysis

Innovations

  • Open-access evaluation framework ACT-Bench for quantifying action fidelity in world models for autonomous driving
  • Baseline world model Terra trained on multiple large-scale trajectory-annotated datasets to improve action fidelity
  • Large-scale dataset pairing short context videos from nuScenes with corresponding future trajectory data for action fidelity evaluation

Methodology

The paper proposes ACT-Bench, an open-access evaluation framework that includes a dataset pairing short context videos from nuScenes with future trajectory data. This dataset serves as conditional input for generating future video frames and enables quantitative evaluation of action fidelity. Additionally, a baseline world model, Terra, is trained on multiple large-scale trajectory-annotated datasets to enhance adherence to action instructions. The framework is used to compare Terra against state-of-the-art models.

Key Results

The state-of-the-art model does not fully adhere to given action instructions, while Terra achieves improved action fidelity.

Resources

PaperCodeProjectHugging Face

Tags