Reproduce “DeRep: Learning Depth Representations for End-to-End Navigation in Cluttered Environments” 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
Hanyu Jin, Kenji Shimada
DeRep trains a depth encoder to reconstruct images and predict a local map of connected reachable space and path distance, then freezes the encoder for navigation. Across tested policy-learning methods, reachability supervision improves results over reconstruction-only baselines. In PPO, DeRep reaches 86.5% success versus AE’s 65.2% on unseen tilted obstacles, but Reach performs best on concave obstacles and severe corruption.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.