Search papers, labs, and topics across Lattice.
This paper introduces NavWM, a unified navigation world model that integrates latent world reasoning, multimodal action prediction, and controllable visual generation to address the limitations of conventional visual navigation policies. By leveraging latent world tokens for geometric and semantic understanding and employing an anchor-based multimodal trajectory forecasting framework, NavWM enhances decision-making in complex environments. Experimental results show that NavWM achieves significant advancements in high-fidelity future state generation and zero-shot navigation success, outperforming existing methods.
NavWM redefines navigation by combining perception, generation, and control into a single framework, leading to unprecedented improvements in planning and foresight.
Conventional visual navigation policies often struggle with myopic decision-making and mode collapse in complex environments. While world models offer a promising alternative, existing paradigms typically isolate perception, generation, and control, failing to capture their shared spatio-temporal dynamics. In this paper, we propose NavWM, a unified navigation world model that seamlessly integrates latent world reasoning, multimodal action prediction, and controllable visual generation. At its core, NavWM leverages latent world tokens to distill geometric and semantic priors, endowing the agent with robust structural understanding. To overcome the limitations of deterministic policies, we introduce an anchor-based multimodal trajectory forecasting framework that generates a diverse action space. This inherent diversity explicitly empowers the generative world model to act as a robust closed-loop planner, utilizing visual foresight to evaluate and select the optimal path. Extensive experiments across diverse robotics datasets demonstrate that NavWM significantly advances the state-of-the-art, delivering remarkable improvements in both high-fidelity future state generation and zero-shot navigation success.
