Latent-Space Autoregressive World Model for Efficient and Robust Image-Goal Navigation

2025-11-14

AI Summary

Traditional visual object navigation relies on high-precision localization and mapping, incurring substantial computational overhead. While latent-space world models show promise, they remain hampered by high training and inference costs. This paper introduces the first fully end-to-end latent-space world model for navigation—lightweight, pixel-reconstruction-free, and explicitly designed for efficiency. Instead of reconstructing pixels, it directly models long-horizon spatiotemporal dependencies via autoregressive multi-frame prediction within a compact latent feature space. The approach integrates action-conditioned sequence prediction with latent-space planning, requiring only an encoder–predictor architecture to enable full-cycle navigation learning. Experiments demonstrate a 3.2× reduction in training time, a 447× decrease in planning latency, a 35% improvement in task success rate, and an 11% gain in Success weighted by Path Length (SPL), jointly achieving significant gains in both efficiency and performance.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Computer Vision: Diffusion Models for Vision
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Traditional navigation methods rely heavily on accurate localization and mapping. In contrast, world models that capture environmental dynamics in latent space have opened up new perspectives for navigation tasks, enabling systems to move beyond traditional multi-module pipelines. However, world model often suffers from high computational costs in both training and inference. To address this, we propose LS-NWM - a lightweight latent space navigation world model that is trained and operates entirely in latent space, compared to the state-of-the-art baseline, our method reduces training time by approximately 3.2x and planning time by about 447x,while further improving navigation performance with a 35% higher SR and an 11% higher SPL. The key idea is that accurate pixel-wise environmental prediction is unnecessary for navigation. Instead, the model predicts future latent states based on current observational features and action inputs, then performs path planning and decision-making within this compact representation, significantly improving computational efficiency. By incorporating an autoregressive multi-frame prediction strategy during training, the model effectively captures long-term spatiotemporal dependencies, thereby enhancing navigation performance in complex scenarios. Experimental results demonstrate that our method achieves state-of-the-art navigation performance while maintaining a substantial efficiency advantage over existing approaches.

Problem

• Reducing computational costs in world model training and inference
• Improving navigation efficiency through latent space prediction and planning
• Enhancing performance in complex scenarios via autoregressive modeling

Innovation

• Lightweight latent space navigation world model
• Autoregressive multi-frame prediction strategy
• Path planning within compact latent representation