This figure holistically illustrates the core ideas behind CompassNav.
The top panel contrasts our End-to-End Goal Navigation paradigm with traditional approaches. Unlike Vision-Language Navigation (VLN), which relies on dense, step-by-step instructions, and complex Modular Navigation pipelines, CompassNav directly maps a high-level goal (e.g., "find the plant") to an action through integrated spatial logical reasoning.
The bottom panel details our core contribution-how to stimulate model reasoning ability: a paradigm shift from "Path Imitation" to "Decision Understanding." While traditional methods train agents to replicate a single expert trajectory and penalize any deviation, our agent learns to evaluate the relative quality of all feasible paths at each decision point. This approach cultivates a true "internal compass," enabling the agent to make more intelligent and flexible decisions in unseen environments.
Abstract
The dominant paradigm for training Large Vision-Language Models (LVLMs) in navigation relies on imitating expert trajectories. This approach reduces the complex navigation task to a sequence-to-sequence replication of a single correct path, fundamentally limiting the agent's ability to explore and generalize. In this work, we argue for and introduce a new paradigm: a shift from Path Imitation to Decision Understanding. The goal of this paradigm is to build agents that do not just follow, but truly understand how to navigate. We materialize this through two core contributions: first, we introduce Compass-Data-22k, a novel 22k-trajectory dataset. Its Reinforcement Fine-Tuning (RFT) subset provides a panoramic view of the decision landscape by annotating all feasible actions with A* geodesic distances. Second, we design a novel gap-aware hybrid reward function that dynamically adapts its feedback to decision certainty, shifting between decisive signals for optimal actions and nuanced scores to encourage exploration. Integrated into an SFT-then-RFT recipe, our CompassNav agent is trained not to memorize static routes, but to develop an internal "compass" that constantly intuits the direction to the goal by evaluating the relative quality of all possible moves. This approach enables our 7B agent to set a new state-of-the-art on Goal navigation benchmarks, outperforming even larger proprietary models, and achieve robust real-world goal navigation on a physical robot.
Overview of our data generation pipeline and Compass-Data-22k dataset
The CompassNav two-stage training pipeline
Real-world Deployment Video
BibTeX
@article{YourPaperKey2024,
title={Your Paper Title Here},
author={First Author and Second Author and Third Author},
journal={Conference/Journal Name},
year={2024},
url={https://your-domain.com/your-project-page}
}