Reproduce “Guaranteed Reach-Avoid for Black-Box Systems through Narrow Gaps via Neural Network Reachability” 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
Long Kiu Chung, Wonsuhk Jung, Srivatsank Pullabhotla, Parth Shinde, Yadu Sunil, Saihari Kota, Luis Felipe Wolf Batista, Cédric Pradalier and 1 more
Researchers at Georgia Institute of Technology's Safe Robotics Lab developed NeuralPARC, a method providing guaranteed reach-avoid capabilities for black-box systems navigating narrow gaps using parameterized policies and neural network reachability. The approach demonstrated provably safe extreme vehicle drift parking and autonomous surface vehicle navigation, outperforming prior methods by effectively managing modeling errors.
Key insights
- ReLU neural networks' equivalence to Piecewise Affine (PWA) systems enables the application of formal reachability analysis to learned black-box system dynamics.
- Incorporating explicitly computed error bounds into the reachability problem by shrinking goals and buffering obstacles provides robust, formal safety guarantees for black-box systems.
- Data-driven learning allows for the exploration of diverse, safe control policies for black-box systems without requiring prior analytical models or nominal trajectories.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.