Stiffness estimation and adaptive control for soft robots

Abstract

Although there has been an astonishing increase in the development of nature- inspired robots equipped with compliant features,i.e.soft robots, their full potential has not been exploited yet. One aspect is that the soft robotics research has mainly focused on their position control only, whilest iffness is managed in open loop. Moreover, due to the difficulties of achieving consistent production of the actuation systems for soft articulated robots and the time-varyingnatureoftheirinternalflexibleelements,whicharesubjecttoplasticdeformation overtime,itiscurrentlyachallengetopreciselydeterminethejointstiffness. . In this regard, the thesis puts an emphasis on stiffness estimation and adaptive control for soft articulated robots driven by antagonistic Variable Stiffness Actuators (VSAs) with the aim to impose the desired dynamics of both position and stiffness, which would finally contribute to the overall safety and improved performance of a soft robot. By building upon Unknown Input Observer (UIO) theory, invasive and non-invasive solutions for estimation of stiffness in pneumatic and electro-mechanical actuators are proposed and in the latter case also experimentally validated. Beyond the linearity and scalability advantage, the approaches have an appealing feature that torque and velocity sensors are not needed. Once the stiffness is determined, innovative control approaches are introduced for soft articulated robots comprising an adaptive compensator and a dynamic decoupler. The solutions are able to cope with uncertainties of the robot dynamic model and, when the desired stiffness is constant or slowly-varying, also of the pneumatic actuator. Their verification is performed via simulations and then the pneumatic one is successfully tested on an experimental setup. Finally, the thesis shows via extensive simulations the effectiveness of adaptive technique ap- plied to soft-bodied robots, previously deriving the sufficient and necessary conditions for the controller convergence.

Iako se danas izuzetno intenzivno radi na razvoju robota inspirisanih prirodom koje odlikuje elastična struktura, njihov puni potencijal jox uvek nije iskorišćen. Sa jedne strane, istraživanja u oblasti popustljivih robota su uglavnom fokusirana samo na upravljanje njihovom pozicijom, dok se krutost reguliše u otvorenoj sprezi. Pored toga, zbog poteškoća u postiznju konzistentne proizvodnje aktuatora i promenljive prirode njihovih elastičnih elemenata, koji su vremenom podlo_ni plastičnoj deformaciji, trenutno je izazov precizno odrediti krutost zglobova robota. U cilju doprinosa poboljšanja_u performansi i bezbednosti rada popustivih robota, teza prikazuje doprinos proceni krutosti i adaptivnog simultanog upravljanja pozicijom i krutosti antagonističkih aktuatora promenljive krutosti (VSA). Oslanjajući se na teoriju opservera nepoznatih ulaza (UIO), predložena su invazivna i neinvazivna rešenja za procenu krutosti u pneumatskim i elektromehaničkim aktuatorima i eksperimentalno verifikovana u slučaju druge grupe aktuatora. Pored linearnosti i skalabilnosti, ovi pristupi imaju privlaqnu osobinu da senzori momenta i brzine nisu potrebni. Teza predla_e inovativne sisteme upravljanja koji poseduju adaptivni kompenzator i dinamički dekupler. Predložene metode upravljanja demonstriraju mogućnost da kompenzuju nesigurnosti dinamičkog modela robota bez obzira da li je on pogođen električnim ili pneumatskim aktuatorima. Nakon simulacija, razvijeno upravljanje je verifikovano i na pneumatskom robotu. Na kraju teze, obimne simulacije pokazuju efikasnost adaptivne tehnike kada se primeni na robote sa fleksibilnim linkovima, prethodno izvodeći dovoljne i potrebne uslove za konvergenciju kontrolera.