MetricNet: Recovering Metric Scale in Generative Navigation Policies

2025-09-17

AI Summary

Generative navigation suffers from two fundamental limitations: (1) trajectory predictions lack metric scale, causing spatial misalignment in physical environments; and (2) control policies target only individual waypoints without considering the global path, leading to myopic behavior and collision risks. To address these issues, we propose MetricNet—the first end-to-end framework for metric-scale recovery in generative navigation—which estimates true Euclidean distances between waypoints via deep regression, thereby aligning pixel/feature-space representations with physical coordinates. Building upon this, we introduce MetricNav, a navigation policy that preserves the expressive power of generative modeling while integrating long-horizon path constraints and local obstacle avoidance. Extensive evaluations on both simulation and real-world robotic platforms demonstrate significant improvements in navigation success rate, exploration efficiency, and collision safety. Our approach establishes a measurable, deployable paradigm for generative navigation.

Technology Category

• Planning, Routing, and Scheduling: Replanning and Plan Repair
• Intelligent Robots: Localization, Mapping, and Navigation
• Natural Language Processing: Generation

Application Category

• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Generative navigation policies have made rapid progress in improving end-to-end learned navigation. Despite their promising results, this paradigm has two structural problems. First, the sampled trajectories exist in an abstract, unscaled space without metric grounding. Second, the control strategy discards the full path, instead moving directly towards a single waypoint. This leads to short-sighted and unsafe actions, moving the robot towards obstacles that a complete and correctly scaled path would circumvent. To address these issues, we propose MetricNet, an effective add-on for generative navigation that predicts the metric distance between waypoints, grounding policy outputs in real-world coordinates. We evaluate our method in simulation with a new benchmarking framework and show that executing MetricNet-scaled waypoints significantly improves both navigation and exploration performance. Beyond simulation, we further validate our approach in real-world experiments. Finally, we propose MetricNav, which integrates MetricNet into a navigation policy to guide the robot away from obstacles while still moving towards the goal.

Problem

• Recovering metric scale in generative navigation policies
• Addressing abstract unscaled trajectory space issues
• Preventing short-sighted unsafe robot navigation actions

Innovation

• MetricNet predicts metric distance between waypoints
• Grounds policy outputs in real-world coordinates
• Integrates obstacle avoidance with goal navigation