Reproduce “Feudal Networks for Visual Navigation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Faith Johnson, Bryan Bo Cao, Ashwin Ashok, Shubham Jain, Kristin Dana

Researchers from Rutgers, Stony Brook, and Georgia State Universities developed FeudalNav, a hierarchical framework for visual navigation that achieves strong performance in image-goal tasks without relying on reinforcement learning, topological graphs, odometry, or metric maps. The system outperforms baselines and is competitive with state-of-the-art methods, particularly on curved trajectories, while utilizing significantly less training data.

Key insights

  • Effective visual navigation can be achieved without traditional components like Reinforcement Learning, odometry, explicit topological graphs, or dense metric maps, challenging long-held assumptions in the field.
  • Hierarchical learning, inspired by feudal principles, simplifies complex navigation tasks by decomposing them into specialized agents operating at different spatial and temporal scales.
  • A self-supervised memory representation combined with imitation learning from human demonstrations enables robust and generalizable navigation in previously unseen environments.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.