Reproduce “ArtPlanner: Robust Legged Robot Navigation in the Field” 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
Lorenz Wellhausen, Marco Hutter
ArtPlanner enables robust, real-time navigation for legged robots by combining geometric reachability, learned foothold scores, and a learned motion cost within a novel batched graph construction framework. This system successfully guided quadrupedal robots through complex, unknown subterranean environments, contributing to a win in the DARPA Subterranean Challenge.
Key insights
- Integrating geometric reachability with learned foothold scores and dynamic motion costs allows legged robots to leverage their unique mobility for precise maneuvering while explicitly assessing locomotion risk.
- Batching neural network queries for graph edge costs significantly reduces computational overhead and enables a bounded computation time, which is critical for continuous real-time autonomous navigation.
- Robust height map pre-processing, including virtual surfaces for occlusions, safety thresholds for dangerous edges, and ceiling point filters, is essential for safe and effective planning in perceptually challenging environments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.