Search papers, labs, and topics across Lattice.
AnyImageNav is introduced as a training-free system for precise last-meter image goal navigation, addressing the limitations of the standard 1m success criterion by treating the goal image as a geometric query. The method uses a semantic relevance signal to guide exploration and trigger a 3D multi-view foundation model for pose recovery when proximity is detected. Results demonstrate state-of-the-art navigation success rates and a 5-10x improvement in pose recovery accuracy compared to adapted baselines on Gibson and HM3D datasets.
Achieve 5-10x more precise robot navigation by treating any target image as a geometric query, enabling accurate 6-DoF pose recovery without any training.
Image Goal Navigation (ImageNav) is evaluated by a coarse success criterion, the agent must stop within 1m of the target, which is sufficient for finding objects but falls short for downstream tasks such as grasping that require precise positioning. We introduce AnyImageNav, a training-free system that pushes ImageNav toward this more demanding setting. Our key insight is that the goal image can be treated as a geometric query: any photo of an object, a hallway, or a room corner can be registered to the agent's observations via dense pixel-level correspondences, enabling recovery of the exact 6-DoF camera pose. Our method realizes this through a semantic-to-geometric cascade: a semantic relevance signal guides exploration and acts as a proximity gate, invoking a 3D multi-view foundation model only when the current view is highly relevant to the goal image; the model then self-certifies its registration in a loop for an accurate recovered pose. Our method sets state-of-the-art navigation success rates on Gibson (93.1%) and HM3D (82.6%), and achieves pose recovery that prior methods do not provide: a position error of 0.27m and heading error of 3.41 degrees on Gibson, and 0.21m / 1.23 degrees on HM3D, a 5-10x improvement over adapted baselines.
