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

UniUGP: Unifying Understanding, Generation, and Planing For End-to-end Autonomous Driving

arXiv 25.12 2025 44.4 method, benchmark, application

TLDR

UniUGP unifies understanding, generation, and planning for autonomous driving via hybrid experts and four-stage training, achieving SOTA on long-tail scenarios.

Reasoning

The paper proposes a novel framework integrating VLMs and video generation for AD, with strong empirical results on multiple datasets. However, the abstract lacks detailed evaluation metrics and may suffer from architectural complexity.

Read-first score

Read-first score 44.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 18.

Recency 8%
86.7

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

Methodology quality 25%
60

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

Reproducibility 25%
46

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

Topical relevance 42%
25.7

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

Frontier

Rank sensitivity

Stability: volatile; rank range: 371.

Keyword Scores

world dynamics prediction
5
video world model
4
generative world model
3
world model
2
model-based reinforcement learning world model
2
world simulator
1
interactive world model
1

Deep Analysis

Innovations

  • Unified Understanding-Generation-Planning framework (UniUGP) that synergizes scene reasoning, future video generation, and trajectory planning through a hybrid expert architecture
  • Four-stage training strategy that progressively builds capabilities across multiple existing AD datasets and newly constructed specialized datasets
  • Construction of multiple specialized datasets providing reasoning and planning annotations for complex scenarios

Methodology

UniUGP integrates pre-trained vision-language models (VLMs) and video generation models into a hybrid expert architecture. It takes multi-frame observations and language instructions as input, and produces interpretable chain-of-thought reasoning, physically consistent trajectories, and coherent future videos. A four-stage training strategy is employed to progressively build these capabilities across multiple existing autonomous driving datasets and the proposed specialized datasets.

Key Results

Experiments demonstrate state-of-the-art performance in perception, reasoning, and decision-making, with superior generalization to challenging long-tail situations.

Tags