Physical human-robot interaction controlof an assistive arm exoskeleton
Authors
Pedersen, Andreas ; Murcia I Matute, Marc
Term
4. term
Education
Publication year
2021
Pages
91
Abstract
Dette speciale omhandler fysisk menneske-robot interaktion for et assistivt arm-exoskelet med en variabel stivheds mekanisme (VSM) og retter sig mod sikker, eftergivelig assistance. Formålet er at udvikle og evaluere en kontrolarkitektur, der dækker to brugsscenarier: når en last er fastgjort til exoskelettet, og når brugeren selv holder lasten. Metodisk opstilles en HRI-model som grundlag for en todelt løsning: en lavniveau positionskontrol baseret på en gain-scheduling LQR-kontroller, der håndterer VSM’ens ikke-linearitet via parameterestimering, samt en højniveau admittans-baseret metode, der bruger en kraftsensor til at forme den følte dynamik og generere positionsreferencer. Stabiliteten undersøges numerisk, og begge kontrolniveauer implementeres på hardware (bl.a. Arduino DUE) med sensorkalibrering og praktiske tests; derudover foreslås en sensorløs strategi i simulationer. Resultaterne viser, at lavniveaukontrollen opnår en gennemsnitlig positionssporingsfejl omkring 6 %, og at systemet kan kompensere for exoskelettets egenvægt ved forskellige belastninger, operere i lavimpedans-tilstand uden aktiv assistance og yde assistiv støtte ved håndholdt last. I den assistive test var nøjagtigheden dog begrænset af usikker lastestimering, hvilket gav omkring 28 % fejl i det tilsigtede hjælpeniveau. Samlet set understøtter de eksperimentelle resultater de foreslåede strategier som et proof of concept og peger på forbedringsmuligheder i lastestimering og positionssporing.
This thesis addresses physical human–robot interaction for an assistive arm exoskeleton with a variable stiffness mechanism (VSM), aiming for safe and compliant assistance. The objective is to develop and evaluate a control architecture for two use cases: when a payload is attached to the exoskeleton and when the user grasps the payload. A human–robot interaction model underpins a two-layer solution: a low-level position controller using gain-scheduled LQR to manage the VSM’s nonlinearity through parameter estimation, and a high-level admittance-based method that uses an interaction force sensor to shape the perceived dynamics and generate position references. Stability is studied numerically, both control layers are implemented on hardware (including an Arduino DUE) with sensor calibration and practical tests, and a sensorless strategy is proposed in simulations. Results show an average position tracking error of about 6% for the low-level controller, and demonstrate robot self-weight compensation across loads, low-impedance following without active assistance, and assistive behavior for a handheld payload. In the assistive case, accuracy is limited by uncertain payload estimation, leading to roughly 28% error in the intended assistance level. Overall, the experiments support the proposed strategies as a proof of concept and point to improvements in payload estimation and position tracking.
[This summary has been generated with the help of AI directly from the project (PDF)]
Documents