World Models for Autonomous Navigation of Terrestrial Robots from LIDAR Observations

JIFS 25 2025 56 method, application

TLDR

A model-based RL framework using DreamerV3 and MLP-VAE world model for LIDAR-based robot navigation achieves 100% success in simulation.

Reasoning

The paper presents a clear contribution: integrating a world model with latent representations for LIDAR data, showing strong simulated results. However, it lacks real-world experiments and does not address generative, interactive, or video world models, limiting the scope of its claims.

Read-first score

Read-first score 56, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 29.

Recency 8%

86.7

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

Methodology quality 25%

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

Reproducibility 25%

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

Topical relevance 42%

41.4

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 401.

Keyword Scores

world model

model-based reinforcement learning world model

world dynamics prediction

world simulator

generative world model

interactive world model

video world model

Deep Analysis

Innovations

Integration of a Multi-Layer Perceptron Variational Autoencoder (MLP-VAE) within the DreamerV3 world model to encode high-dimensional LIDAR readings into compact latent representations
Use of learned latent dynamics and imagination-based policy optimization for autonomous navigation from full-resolution LIDAR observations
Demonstration of 100% success rate on simulated TurtleBot3 navigation tasks, significantly outperforming model-free baselines (SAC, DDPG, TD3) which plateau below 85%

Methodology

The proposed framework builds on DreamerV3, a model-based RL algorithm, and incorporates an MLP-VAE to compress 360-dimensional LIDAR readings into a compact latent space. A learned dynamics predictor then models transitions in this latent space, enabling efficient policy optimization through imagined trajectories without requiring full-resolution sensor data at each step.

Key Results

The DreamerV3-based agent achieves a 100% success rate across all evaluated simulated environments using full LIDAR data, while model-free baselines (SAC, DDPG, TD3) plateau below 85%. The proposed architecture also demonstrates faster convergence compared to these baselines.

Limitations

Only evaluated in simulation on TurtleBot3, not on real-world robots
Relies solely on LIDAR observations; generalization to other sensor modalities or multi-sensor fusion is not addressed
Limited to a single robot platform and specific navigation tasks; broader applicability remains unverified

Resources

Paper