Effective robot navigation in unseen environments is a challenging task that requires precise control actions at high frequencies. Recent advances have framed it as an image-goal-conditioned control problem, where the robot generates navigation actions using frontal RGB images. Current state-of-the-art methods in this area use diffusion policies to generate these control actions. Despite their promising results, these models are computationally expensive and suffer from weak perception. To address these limitations, we present FlowNav, a novel approach that uses a combination of conditional flow matching and depth priors from off-the-shelf foundation models to learn action policies for robot navigation. FlowNav is significantly more accurate at navigation and exploration than state-of-the-art methods. We validate our contributions using real robot experiments in multiple unseen environments, demonstrating improved navigation reliability and accuracy.
FlowNav is able to navigate across multiple obstacle configurations, even identifying glass walls.
FlowNav demonstrates that including depth priors during model training helps the model more representative better action policies for robot navigation.
Flow matching policies remain effective with fewer inference steps, thus reducing trajectory generation time.
@misc{gode2025flownav,
title={FlowNav: Combining Flow Matching and Depth Priors for Efficient Navigation},
author={Gode, Samiran and Nayak, Abhijeet and Oliveira, Débora N.P. and and Krawez, Michael
and Schmid, Cordelia and Burgard, Wolfram},
year={2025},
eprint={2411.09524},
archivePrefix={arXiv},
primaryClass={cs.RO},
url={https://arxiv.org/abs/2411.09524},
}