Reproduce “Socially Acceptable Bipedal Robot Navigation via Social Zonotope Network Model Predictive Control” 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
Abdulaziz Shamsah, Krishanu Agarwal, Nigam Katta, Abirath Raju, Shreyas Kousik, Ye Zhao
The Georgia Tech research team developed a Social Zonotope Network Model Predictive Control (SZN-MPC) framework that enables bipedal robots to navigate dynamic, human-populated environments safely and acceptably. It learns social behaviors and robustly plans paths using zonotopes, validated through real-time implementation on a Digit bipedal robot.
Key insights
- Learning socially acceptable behaviors directly as reachable sets (zonotopes) within a neural network enables efficient and robust integration of social intelligence and safety guarantees into robot planning.
- Embedding bipedal robot dynamic constraints (e.g., velocity limits, heading changes) via STL losses during neural network training ensures that the learned social paths are inherently executable and safe for complex legged systems.
- Online zonotope refinements for personal space and modeling error compensation are crucial for practical hardware deployment, bridging the gap between learned behaviors in simulation and robust performance in real-world scenarios.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.