2020/04/03 by Carl Folkestad, Folkestad, Carl, D. Pastor +4 · 2 citations
Computer Science · Engineering · Mathematics · Physics and Astronomy · #Aerospace engineering #Artificial intelligence #Computer science #Control (management) #Control theory (sociology) #Diffeomorphism #Engineering #FOS: Electrical engineering #Flow (mathematics) #Fuel Cells and Related Materials #Mathematical analysis #Mathematics #Microfluidic and Capillary Electrophoresis Applications #Model Reduction and Neural Networks #Multirotor #Nonlinear system #Physics #State space #Systems and Control (eess.SY) #cs.SY #eess.SY #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2004.01708
published in arXiv (Cornell University) (Cornell University) · Accepted to the International Conference on Robotics and Automation 2020 (ICRA). arXiv admin note: text overlap with arXiv:1911.08751
arxiv created 2020/04/03 · openalex publication_date 2020/04/03 · arxiv updated 2020/04/07 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/05
This paper presents a novel episodic method to learn a robot's nonlinear\ndynamics model and an increasingly optimal control sequence for a set of tasks.\nThe method is based on the em Koopman operator approach to nonlinear\ndynamical systems analysis, which models the flow of em observables in a\nfunction space, rather than a flow in a state space. Practically, this method\nestimates a nonlinear diffeomorphism that lifts the dynamics to a higher\ndimensional space where they are linear. Efficient Model Predictive Control\nmethods can then be applied to the lifted model. This approach allows for real\ntime implementation in on-board hardware, with rigorous incorporation of both\ninput and state constraints during learning. We demonstrate the method in a\nreal-time implementation of fast multirotor landing, where the nonlinear ground\neffect is learned and used to improve landing speed and quality.\n