Latent-WAM: Latent World Action Modeling for End-to-End Autonomous Driving

arXiv 26.3 2026 61.6 method, application

TLDR

Latent-WAM uses spatially-aware and dynamics-informed latent world representations for efficient end-to-end autonomous driving, achieving SOTA on NAVSIM v2 and HUGSIM.

Reasoning

The paper introduces novel modules (SCWE and DLWM) that effectively address key limitations of prior world-model planners, with strong empirical results on two benchmarks using a compact model. However, the evaluation is limited to simulated environments, and real-world validation is absent.

Read-first score

Read-first score 61.6, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 40.

Recency 8%

100

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

Methodology quality 25%

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

Topical relevance 42%

57.1

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

Reproducibility 25%

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 290.

Keyword Scores

world model

world dynamics prediction

generative world model

video world model

world simulator

model-based reinforcement learning world model

interactive world model

Deep Analysis

Innovations

Spatial-Aware Compressive World Encoder (SCWE) that distills geometric knowledge from a foundation model and compresses multi-view images into compact scene tokens via learnable queries
Dynamic Latent World Model (DLWM) that employs a causal Transformer to autoregressively predict future world status conditioned on historical visual and motion representations
Achieves strong trajectory planning through spatially-aware and dynamics-informed latent world representations, addressing inadequately compressed representations, limited spatial understanding, and underutilized temporal dynamics

Methodology

Latent-WAM consists of two core modules: a Spatial-Aware Compressive World Encoder (SCWE) that uses a foundation model to extract geometric knowledge and compresses multi-view images into compact scene tokens via learnable queries, and a Dynamic Latent World Model (DLWM) that employs a causal Transformer to autoregressively predict future world states conditioned on historical visual and motion representations. The framework is trained end-to-end for trajectory planning, evaluated on NAVSIM v2 and HUGSIM benchmarks against prior perception-free methods.

Key Results

Latent-WAM achieves state-of-the-art results with 89.3 EPDMS on NAVSIM v2 and 28.9 HD-Score on HUGSIM, surpassing the best prior perception-free method by 3.2 EPDMS while using significantly less training data and a compact 104M-parameter model.

Resources

Paper