Reproduce “Provably-Safe, Online System Identification” 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
Bohao Zhang, Zichang Zhou, Ram Vasudevan
This paper presents a framework for provably-safe, online system identification that allows robotic manipulators to autonomously identify unknown payload parameters while rigorously adhering to safety constraints. The method successfully performs challenging manipulation tasks on a Kinova-gen3 robot with various dumbbells, demonstrating faster and more accurate parameter convergence with tighter uncertainty bounds compared to baselines.
Key insights
- Integrating robust motion planning (ARMOUR) with system identification enables the generation of trajectories that are both provably safe and optimally exciting for parameter estimation.
- Momentum-based dynamics formulation and Log-Cholesky parameterization allow for accurate and physically consistent estimation of inertial parameters without noisy acceleration measurements, and transform constrained optimization into an unconstrained one.
- Propagating bounded sensor noise and initial uncertainties through a perturbation analysis yields provably overapproximative interval bounds for identified parameters, providing robust uncertainty quantification.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.