Vision-Centric 4D Occupancy Forecasting and Planning via Implicit Residual World Models

arXiv 25.10 2025 49.8 method, application

TLDR

Proposes IR-WM, an implicit residual world model for 4D occupancy forecasting and planning, achieving top performance on nuScenes.

Reasoning

The paper introduces a novel residual prediction approach to avoid redundant static background modeling, supported by strong empirical results on nuScenes. Strengths include the alignment module and coupling schemes for planning; weaknesses include limited evaluation to a single benchmark and potential lack of generalizability.

Read-first score

Read-first score 49.8, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 38.

Recency 8%

86.7

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

Topical relevance 42%

54.3

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

Methodology quality 25%

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

Reproducibility 25%

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

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 464.

Keyword Scores

world model

world dynamics prediction

generative world model

model-based reinforcement learning world model

video world model

interactive world model

world simulator

Deep Analysis

Innovations

Implicit Residual World Model (IR-WM) that predicts only the residual changes instead of full future scene reconstruction
Leveraging BEV features from previous timestep as a strong temporal prior for residual prediction
Alignment module to calibrate semantic and dynamic misalignments to reduce error accumulation
Investigation of different forecasting-planning coupling schemes demonstrating that implicit future state improves planning accuracy

Methodology

IR-WM first builds a bird's-eye-view (BEV) representation of the current state from visual observations. It then uses BEV features from the previous timestep as a temporal prior and predicts only the residual changes conditioned on ego-vehicle actions and scene context. An alignment module is applied to calibrate semantic and dynamic misalignments to alleviate error accumulation over time.

Key Results

On the nuScenes benchmark, IR-WM achieves top performance in both 4D occupancy forecasting and trajectory planning.

Resources

Paper