Reproduce “Target-driven Visual Navigation in Indoor Scenes using Deep Reinforcement Learning” 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

Yuke Zhu, Roozbeh Mottaghi, Eric Kolve, Joseph J. Lim, Abhinav Gupta, Li Fei-Fei, Ali Farhadi

The authors present a goal-conditioned navigation policy using the robot’s current view and a target image specifying location and viewpoint. It shares representations across goals, with scene-specific layers, and is trained in AI2-THOR. After 100 million frames, it averages 210.7 steps versus 723.5 for A3C. Held-out-target success depends on target count and proximity; transfer involved fine-tuning in a small room.

Summary by alphaXiv.

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