Naslov
Koopman-Based Modeling and Control of Nonlinear Soft Robots

Autori
Kamenar, Ervin ; Haggerty, David ; Banks, Michael ; Cao, Alan ; Mezić, Igor ; Hawkes, Elliot

Vrsta, podvrsta i kategorija rada
Sažeci sa skupova, neobjavljeni rad, znanstveni

Skup
2023 SIAM Conference on Dynamical Systems

Mjesto i datum
Portland (OR), Sjedinjene Američke Države, 14.05.2023. - 19.05.2023

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Recenziran

Ključne riječi
Koopman operator ; soft robotics ; data-driven modeling and control

Sažetak
Compared to traditional, rigid robots, soft robots offer inherent compliance that gives these robots the ability to work safely in close proximity with humans. However, the flexibility of soft robots induces highly nonlinear dynamical behavior which makes them very difficult to model and control. Considering also the infinite number degrees-of-freedom present in such systems, obtaining mathematical models grounded on first-principles and knowledge of material properties is very difficult and generally enables their operating only in linear regime (low deflection and low velocity). To address the above listed issues, a data-driven learning approach based on the Koopman operator framework is used to identify the model of pneumatically driven soft arm. The arm is equipped with a motion capture system to measure its position in workspace. An Extended Dynamic Mode Decomposition (EDMD) with time delay observables is applied to obtain a finite dimensional approximation of control Koopman operator. Such approach results in a globally linear model of the nonlinear soft robot dynamics using only a couple of minutes of training data generated by random step inputs. Based on the obtained linear model, a simple linear state feedback controller is developed and precise, fast, high-deflection displacements to arbitrary reference positions commanded in real-time are achieved.

Izvorni jezik
Engleski

Znanstvena područja
Strojarstvo, Interdisciplinarne tehničke znanosti

POVEZANOST RADA