Xiaomi EV World Model: A Joint World Model Integrating Reconstruction and Generation for Autonomous Driving

arXiv 26.5 2026 49 method, application

TLDR

A joint world model for autonomous driving integrating reconstruction (WorldRec) and generation (WorldGen) for improved stability and fidelity.

Reasoning

The paper presents a unified system combining reconstruction and generation for world models in autonomous driving, with novel techniques like sparse scene queries and two-stage training. However, the abstract lacks explicit real-world evaluation or comparison to existing methods, and the scope is limited to autonomous driving.

Read-first score

Read-first score 49, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 43.

Recency 6%

100

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

Citation impact 18%

71.9

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.71878087

Topical relevance 29%

61.4

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=none

Reproducibility 18%

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

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: 339.

Keyword Scores

world model

generative world model

video world model

world dynamics prediction

world simulator

model-based reinforcement learning world model

interactive world model

Deep Analysis

Innovations

WorldRec: a feed-forward reconstruction architecture driven by sparse scene queries that initializes structured queries in 3D space to aggregate cross-view, cross-temporal features, enforcing spatial consistency and yielding compact 3D Gaussian scene representations.
WorldGen: a two-stage training framework of bidirectional pretraining followed by causal fine-tuning through three progressive stages (Teacher Forcing, ODE distillation, and DMD), enabling high-quality online causal video generation in as few as 4 denoising steps.
JWM: a joint world model that deeply integrates WorldRec and WorldGen to achieve synergistic gains in generation stability, cross-frame consistency, and visual fidelity for autonomous driving applications.

Methodology

The paper proposes a unified technical system with two core modules: WorldRec for world representation using sparse 3D queries to aggregate cross-view and cross-temporal features into compact 3D Gaussian scene representations, and WorldGen for world generation using a two-stage training framework (bidirectional pretraining then causal fine-tuning via Teacher Forcing, ODE distillation, and DMD) for efficient online causal video generation. The Joint World Model (JWM) integrates both modules to leverage their complementary strengths.

Key Results

The JWM achieves synergistic gains in generation stability, cross-frame consistency, and visual fidelity, enabling high-quality online causal video generation in as few as 4 denoising steps, providing a solid foundation for closed-loop simulation, data synthesis, and end-to-end training in autonomous driving.

Resources

Paper