Reproduce “AirSplan: Risk-Aware Motion Planning for Quadrotors in Cluttered 3D Gaussian Splats” 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
Seth Isaacson, William Hong, Katherine A. Skinner, Ram Vasudevan
AirSplan plans quadrotor trajectories through cluttered scenes reconstructed as normalized 3D Gaussian Splats, combining continuous-time reachability bounds with a detailed sphere-based model of the vehicle. In 500 simulated trials, it reached goals collision-free in 81.2% of cases, compared with 51.2% for Splat-Nav and 28.8% for CATNIPS; its BVH-accelerated collision checks averaged 42.2 ms.
Key insights
- Representing the aircraft with multiple spheres and accounting for orientation-dependent motion gives a less restrictive swept-volume bound than treating the whole quadrotor as one sphere.
- Continuous-time reachability and probabilistic scene occupancy can be combined for risk-aware planning, while spatial indexing reduces the cost of evaluating many robot spheres against large Gaussian scenes.
- The full system’s performance depends on both its reachability representation and its sampling-based optimizer; results are from simulation and assume an accurate scene representation and trajectory tracking.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.