EA-Nav: Learning Safe Visual Navigation Policies with Embodiment Awareness

2026-07-22

AI Summary

This work addresses the challenge of action ambiguity arising from morphological differences among embodied agents when navigating using only visual observations. To tackle this issue, the authors propose an imitation learning–based embodied perception framework that explicitly incorporates embodied geometry into the learning process. The approach features a multi-stage design: during pretraining, a cross-embodiment navigation dataset is constructed from internet videos, with embodied geometry encoded as conditional tokens; during fine-tuning, a decoupled architecture leverages trajectory augmentation to generate high-risk samples, enabling separate optimization of spatial perception and risk correction modules. Experimental results demonstrate that the proposed method significantly improves navigation performance and safety in cross-embodiment settings, underscoring the critical role of embodied geometry in mitigating action ambiguity and enhancing generalization.

Abstract

Cross-embodiment navigation is a key challenge in embodied intelligence. Due to differences in embodiment, the same visual observation may imply different actions for different agents, making prediction ambiguous when relying solely on vision. Existing studies mainly rely on reinforcement learning, which requires large-scale interaction and careful reward design, making it difficult to support scalable pretraining and real-world adaptation. In contrast, imitation-learning-based approaches remain limited. To address these challenges, we propose an imitation-learning-based embodiment-aware navigation framework with a modular multi-stage design. In pretraining, we construct a cross-embodiment navigation dataset from Internet videos and introduce embodiment geometry as conditional tokens to reduce action ambiguity under the same observation. In fine-tuning, we design a multimodal information injection mechanism based on a decoupled architecture. Specifically, we design a trajectory augmentation strategy to generate high-risk samples, which are used to train spatial perception and risk-aware correction separately, thereby explicitly incorporating embodiment geometry for safe navigation. Experimental results show that the proposed method effectively improves navigation performance across different embodiment settings, demonstrating the effectiveness of incorporating embodiment geometry into embodied navigation.

Problem

• cross-embodiment navigation
• embodiment awareness
• visual navigation
• action ambiguity
• embodied intelligence

Innovation

• embodiment-aware navigation
• imitation learning
• cross-embodiment
• geometric conditioning
• risk-aware correction