EvolveNav: Proactive Preflection and Self-Evolving Memory for Zero-Shot Object Goal Navigation

2026-06-16

AI Summary

This work addresses the inefficiency and repetitive errors in zero-shot object navigation caused by agents’ inability to adapt online. To overcome this limitation, the authors propose a self-evolving navigation framework that constructs a memory bank of executable rules distilled from historical trajectories. During inference, the agent dynamically retrieves effective rules using an upper confidence bound–based strategy that balances semantic relevance and past success rates. A novel memory-guided pre-reflection module is introduced to anticipate action outcomes and suppress unproductive exploration. This approach enables continuous self-improvement during testing without additional training, achieving a 10.1% absolute gain in success rate under zero-shot conditions and significantly reducing redundant exploration steps.

Abstract

Zero-Shot Object-Goal Navigation (ZS-OGN) requires embodied agents to explore and locate target objects without any prior training. To this end, recent methods leverage foundation models. But they typically rely on static priors and lack adaptation, which leads to repeated errors and costly trial and error. In this paper, we propose a self-evolving ZS-OGN framework that enables continuous test-time improvement. Specifically, we build an agentic rule memory by extracting actionable knowledge from past trajectories. Then, we propose a retrieval strategy based on upper confidence bound, selecting effective rules by balancing semantic relevance and historical success. In addition, we introduce a memory-guided preflection module that forecasts potential outcomes before action, reducing inefficient exploration. Extensive experiments show that our method outperforms existing zero-shot baselines, achieving a 10.1\% improvement in success rate with fewer unnecessary steps.

Problem

• Zero-Shot Object-Goal Navigation
• embodied agents
• foundation models
• adaptation
• trial and error

Innovation

• self-evolving memory
• zero-shot object-goal navigation
• preflection
• upper confidence bound
• embodied AI