Reproduce “Learning Vision-Based Agile Gap Traversal: Differentiable Simulation with a Warm-Started Critic” 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
Nuthasith Gerdpratoom, Tianchen Sun, Yichao Gao, Lin Zhao
The method trains a geometry-informed flight policy and critic, then transfers and adapts the critic to guide a visual policy using gate masks and body measurements. It reports a peak simulation success rate of 0.893 and real quadrotor trials at four gate tilts with an average maximum speed of 3.5 m/s. Those trials used motion-capture-generated masks, not onboard camera perception.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.