Fast Setup of Robotic Material Processing with Quality Guarantees using Set Invariance

Yitaek Kim

doi:10.21996/fnxf-fe83

Fast Setup of Robotic Material Processing with Quality Guarantees using Set Invariance

Yitaek Kim^*

^*Kontaktforfatter

Publikation: Afhandling › Ph.d.-afhandling

27 Downloads (Pure)

Abstract

Denne afhandling præsenterer en omfattende pipeline til hurtig implementering af robotter til produktionsopgaver af høj kvalitet. Afhandlingen fokuserer på en bestemt anvendelse, navnlig overfladebehandling ved polering. En udbredt metode til hurtig opsætning af robotter er at overføre menneskelige færdigheder til en robot ved hjælp af programmering ved demonstration (PbD, Programming by Demonstration). De nuværende metoder er dog hverken pålidelige eller fleksible nok til at sikre god ydeevne og generalisere kraftrelevante opgaver.

For at løse disse udfordringer præsenterer denne afhandling en modelbaseret platform for overførsel af færdigheder til robotsystemet. Som et eksempel på robotproduktionsprocessen udvikler denne afhandling et robotpoleringssystem baseret på menneskelige demonstrationer. I denne forbindelse defineres menneskelige poleringsfærdigheder ved brug af materialefjernelseshastighed (MRR, Material Removal Rate) og materialefjernelse (MR, Material Removal) baseret på Preston-ligningen. MRR-/MR-profiler beregnes for menneskelig demonstration og overføres til robotten via en MRR-/MR-regulator i robotten. Slutteligt demonstreres det hvorledes en lært færdighed kan generaliseres til objekter med nye geometrier.

Hvis der forekommer kinematiske fejl mellem et givet emne og en robot, kan systemet dog ikke fuldstændig garantere processens kvalitet. For at afhjælpe dette problem præsenterer denne afhandling en kontaktbaseret metode til kinematisk kalibrering af emner. De fleste metoder til kinematisk kalibrering anvender kameraer, og kræver derfor specialviden for at konfigurere det. I modsætning til denne tilgang præsenterer denne afhandling en nem kinæstetisk kalibreringsmetode, der gør det muligt for brugerne at gribe og flytte en robot hen på emnet for at indsamle data og derefter beregne emnets positur. Metoden estimerer i første omgang overfladenormaler og kontaktpunkter, og finder derefter mulige sammenfald mellem punkter på overfladen af 3D CAD-modellen og det estimerede data. Ved at benytte Graduated Non-Convexity (GNC) til robust optimering beregnes emnets positur, hvilket reducerer den kinematiske unøjagtigheder.

Afslutningsvis drøfter denne afhandling sikkerhedsfilteret i slutningen af pipelinen. Til dette formål udnyttes kontrol-barriere-funktioner til at garantere sikkerhed; for eksempel at kontaktkræfter holdes indenfor givne begrænsninger, der sikrer kvaliteten at en robotoverfladepolering. Da der forekommer modelusikkerheder i robotsystemer, præsenterer denne afhandling en ny teoretisk metode til at kombinere den Gaussiske proces (GP, Gaussian Process) med robuste, adaptive kontrol-barriere-funktioner (RaCBF’er, Robust adaptive Control Barrier Functions) for at reducere usikkerheden. Der kan forekomme en ukendt inputforsinkelse i robotsystemet, hvilket kan medføre overtrædelse af sikkerhedsbestemmelser. Derfor præsenterer denne afhandling en metode til at sikre disse i tilfælde af inputforsinkelse. Metoden estimerer den ukendte inputforsinkelse og anvender denne til at forudsige systemmodellens fremtidige tilstand. Ved at inkludere begrænsningerne for beregnings- og tilstandsestimeringsfejl kan robustheden i CBF’ernes sikkerhedstilstand håndhæves.

Originalsprog	Engelsk
Bevilgende institution	Syddansk Universitet
Vejledere/rådgivere	Sloth, Christoffer, Hovedvejleder Petersen, Henrik Gordon, Bivejleder
Dato for forsvar	6. maj 2024
Udgiver	Syddansk Universitet. Det Tekniske Fakultet
DOI	https://doi.org/10.21996/fnxf-fe83
Status	Udgivet - 23. apr. 2024

Dokumenter og links

10.21996/fnxf-fe83

Open Access VersionForlagets udgivne version, 23,3 MB

SDU link

Tjek adgangen til materialet i Syddansk Universitetsbiblioteks søgemaskine

4 Konferencebidrag i proceedings
1 Tidsskriftartikel

Robust Safety-Critical Control for Input-Delayed System with Delay Estimation
Kim, Y., Kim, J., Ames, A. & Sloth, C., 2024, 2024 European Control Conference (ECC). IEEE, s. 2218-2223
Publikation: Kapitel i bog/rapport/konference-proceeding › Konferencebidrag i proceedings › Forskning › peer review
Safe robust adaptive control under both parametric and nonparametric uncertainty
Kim, Y., Iturrate, I., Langaa, J. & Sloth, C., 2024, I: Advanced Robotics. 38, 5, s. 357-366
Publikation: Bidrag til tidsskrift › Tidsskriftartikel › Forskning › peer review

Åben adgang
Fil
38 Downloads (Pure)
Contact-Based Pose Estimation of Workpieces for Robotic Setups
Kim, Y., Kramberger, A., Buch, A. G. & Sloth, C., 2023, 2023 IEEE International Conference on Robotics and Automation (ICRA). IEEE, s. 12324-12330 (Proceedings - IEEE International Conference on Robotics and Automation, Bind 2023-May).
Publikation: Kapitel i bog/rapport/konference-proceeding › Konferencebidrag i proceedings › Forskning › peer review

Citationsformater

@misc{0715ae0071484da3adf062241a49d87f,

title = "Fast Setup of Robotic Material Processing with Quality Guarantees using Set Invariance",

abstract = "This thesis proposes a comprehensive pipeline for the fast deployment of robots for high-quality manufacturing processing tasks. The thesis focuses on surface treatment applications as an example of the process. A popular way to achieve a fast setup is to transfer human skills to a robot by leveraging Programming by Demonstration (PbD). However, the current works are not reliable and flexible enough to ensure good performance and to generalize force-relevant tasks.To remedy the above-mentioned challenges, this thesis proposes a model-based skill transfer framework for the robotic system. As an example of the robotic manufacturing process, this thesis develops the robotic polishing system based on human demonstrations.To this end, human polishing skills are encoded using Material Removal Rate (MRR) and Material Removal (MR) based on Preston's equation. MRR/MR profiles are estimated during human demonstration and transferred to the robot through an MRR/MR tracking controller in the robot. This framework is also generalized into a new geometry, showing a successful skill transfer performance. However, if there exist kinematic errors between a given workpiece and a robot, the system cannot guarantee the quality of the process perfectly.To resolve this issue, this thesis proposes the contact-based workpiece calibration method for robotic automation systems. Most workpiece calibration methods use vision devices, so they require advanced knowledge to set up the calibration system. On the other hand, this thesis proposes an easy and kinesthetic calibration method that lets users grab and move a robot onto the workpiece to collect the data and then estimates the pose of it. The method first estimates the collected contact data that includes the surface normal and contact point and then finds putative correspondences between 3D CAD model and the estimated data. By leveraging a robust non-minimal solver with Graduated Non-Convexity (GNC), the pose of the workpiece is estimated relative to the robot and used to reduce the kinematic discrepancy.Lastly, this thesis considers the safety filter at the end of the pipeline. To this end, we leverage Control Barrier Functions (CBFs) to ensure safety guarantees, for example, contact force bounds and use these bounds to provide quality guarantees in robotic surface polishing. However, since there exists model uncertainty in robotic systems, this thesis proposes a novel theoretical method for combining Gaussian Process (GP) with Robust adaptive Control Barrier Functions (RaCBFs) to reduce the uncertainty. Furthermore, since an unknown input delay might occur in the robotic system, which leads to safety violations, this thesis proposes a method for ensuring safety guarantees in the presence of input delay. The method estimates the unknown input delay and uses the estimated input delay to predict the future state with the system model. With the bounds of estimation and state prediction errors, robustness is enforced in the safety condition of CBFs.",

keywords = "Robotic material processing, Robot safety, Set invariance, Control Barrier Function",

author = "Yitaek Kim",

year = "2024",

month = apr,

day = "23",

doi = "10.21996/fnxf-fe83",

language = "English",

publisher = "Syddansk Universitet. Det Tekniske Fakultet",

address = "Denmark",

school = "SDU",

}

TY - GEN

T1 - Fast Setup of Robotic Material Processing with Quality Guarantees using Set Invariance

AU - Kim, Yitaek

PY - 2024/4/23

Y1 - 2024/4/23

N2 - This thesis proposes a comprehensive pipeline for the fast deployment of robots for high-quality manufacturing processing tasks. The thesis focuses on surface treatment applications as an example of the process. A popular way to achieve a fast setup is to transfer human skills to a robot by leveraging Programming by Demonstration (PbD). However, the current works are not reliable and flexible enough to ensure good performance and to generalize force-relevant tasks.To remedy the above-mentioned challenges, this thesis proposes a model-based skill transfer framework for the robotic system. As an example of the robotic manufacturing process, this thesis develops the robotic polishing system based on human demonstrations.To this end, human polishing skills are encoded using Material Removal Rate (MRR) and Material Removal (MR) based on Preston's equation. MRR/MR profiles are estimated during human demonstration and transferred to the robot through an MRR/MR tracking controller in the robot. This framework is also generalized into a new geometry, showing a successful skill transfer performance. However, if there exist kinematic errors between a given workpiece and a robot, the system cannot guarantee the quality of the process perfectly.To resolve this issue, this thesis proposes the contact-based workpiece calibration method for robotic automation systems. Most workpiece calibration methods use vision devices, so they require advanced knowledge to set up the calibration system. On the other hand, this thesis proposes an easy and kinesthetic calibration method that lets users grab and move a robot onto the workpiece to collect the data and then estimates the pose of it. The method first estimates the collected contact data that includes the surface normal and contact point and then finds putative correspondences between 3D CAD model and the estimated data. By leveraging a robust non-minimal solver with Graduated Non-Convexity (GNC), the pose of the workpiece is estimated relative to the robot and used to reduce the kinematic discrepancy.Lastly, this thesis considers the safety filter at the end of the pipeline. To this end, we leverage Control Barrier Functions (CBFs) to ensure safety guarantees, for example, contact force bounds and use these bounds to provide quality guarantees in robotic surface polishing. However, since there exists model uncertainty in robotic systems, this thesis proposes a novel theoretical method for combining Gaussian Process (GP) with Robust adaptive Control Barrier Functions (RaCBFs) to reduce the uncertainty. Furthermore, since an unknown input delay might occur in the robotic system, which leads to safety violations, this thesis proposes a method for ensuring safety guarantees in the presence of input delay. The method estimates the unknown input delay and uses the estimated input delay to predict the future state with the system model. With the bounds of estimation and state prediction errors, robustness is enforced in the safety condition of CBFs.

AB - This thesis proposes a comprehensive pipeline for the fast deployment of robots for high-quality manufacturing processing tasks. The thesis focuses on surface treatment applications as an example of the process. A popular way to achieve a fast setup is to transfer human skills to a robot by leveraging Programming by Demonstration (PbD). However, the current works are not reliable and flexible enough to ensure good performance and to generalize force-relevant tasks.To remedy the above-mentioned challenges, this thesis proposes a model-based skill transfer framework for the robotic system. As an example of the robotic manufacturing process, this thesis develops the robotic polishing system based on human demonstrations.To this end, human polishing skills are encoded using Material Removal Rate (MRR) and Material Removal (MR) based on Preston's equation. MRR/MR profiles are estimated during human demonstration and transferred to the robot through an MRR/MR tracking controller in the robot. This framework is also generalized into a new geometry, showing a successful skill transfer performance. However, if there exist kinematic errors between a given workpiece and a robot, the system cannot guarantee the quality of the process perfectly.To resolve this issue, this thesis proposes the contact-based workpiece calibration method for robotic automation systems. Most workpiece calibration methods use vision devices, so they require advanced knowledge to set up the calibration system. On the other hand, this thesis proposes an easy and kinesthetic calibration method that lets users grab and move a robot onto the workpiece to collect the data and then estimates the pose of it. The method first estimates the collected contact data that includes the surface normal and contact point and then finds putative correspondences between 3D CAD model and the estimated data. By leveraging a robust non-minimal solver with Graduated Non-Convexity (GNC), the pose of the workpiece is estimated relative to the robot and used to reduce the kinematic discrepancy.Lastly, this thesis considers the safety filter at the end of the pipeline. To this end, we leverage Control Barrier Functions (CBFs) to ensure safety guarantees, for example, contact force bounds and use these bounds to provide quality guarantees in robotic surface polishing. However, since there exists model uncertainty in robotic systems, this thesis proposes a novel theoretical method for combining Gaussian Process (GP) with Robust adaptive Control Barrier Functions (RaCBFs) to reduce the uncertainty. Furthermore, since an unknown input delay might occur in the robotic system, which leads to safety violations, this thesis proposes a method for ensuring safety guarantees in the presence of input delay. The method estimates the unknown input delay and uses the estimated input delay to predict the future state with the system model. With the bounds of estimation and state prediction errors, robustness is enforced in the safety condition of CBFs.

KW - Robotic material processing

KW - Robot safety

KW - Set invariance

KW - Control Barrier Function

U2 - 10.21996/fnxf-fe83

DO - 10.21996/fnxf-fe83

M3 - Ph.D. thesis

PB - Syddansk Universitet. Det Tekniske Fakultet

ER -

Fast Setup of Robotic Material Processing with Quality Guarantees using Set Invariance

Abstract

Dokumenter og links

SDU link

Fingeraftryk

Relaterede publikationer

Robust Safety-Critical Control for Input-Delayed System with Delay Estimation

Safe robust adaptive control under both parametric and nonparametric uncertainty

Contact-Based Pose Estimation of Workpieces for Robotic Setups

Citationsformater