BEGIN:VCALENDAR
PRODID:-//eluceo/ical//2.0/EN
VERSION:2.0
CALSCALE:GREGORIAN
BEGIN:VEVENT
UID:41a774352eb93264cfc86cb3bef8f4a0
DTSTAMP:20260717T075023Z
SUMMARY:Faseeh Ahmad's PhD defence
DESCRIPTION:Kontakt: faseeh.ahmad@cs.lth.se\n\nThesis title:&nbsp\;Towards 
 Self-Reliant Robots: Skill Learning\, Failure Recovery\, and Real-Time Ada
 ptationAuthor: Faseeh Ahmad\, Department of Computer Science\, Lund Univer
 sityFaculty opponent: Professor Lazaros Nalpantidis\, Technical University
  of Denmark (DTU)\, DenmarkExamination Committee:Associate Professor Caspe
 r Schou\, Aalborg University\, DenmarkProfessor Karinne Ramirez-Amaro\, Ch
 almers University of TechnologyDr. Mikael Norrlöf\, ABB Robotics R&amp\;D
  VästeråsDeputy: Senior Lecturer Yiannis Karayiannidis\, Lund University
 Session chair:&nbsp\;Senior Lecturer Michael Doggett\, Lund UniversitySupe
 rvisors:Professor Volker Krueger\, Lund UniversityProfessor Jacek Malec\, 
 Lund UniversityLocation: E:1406\,&nbsp\;E-huset\, Klas Anshelms väg 10/Ol
 e Römers väg 3\, LundHere is a link to download the thesis at LU Researc
 h Portal&nbsp\;AbstractRobots operating in real-world settings must manage
  task variability\, environmental uncertainty\, and failures during execut
 ion. This thesis presents a unified framework for building self-reliant ro
 botic systems by integrating symbolic planning\, reinforcement learning\, 
 behavior trees (BTs)\, and vision-language models (VLMs).At the core of th
 e approach is an interpretable policy representation based on behavior tre
 es and motion generators (BTMGs)\, supporting both manual design and autom
 ated parameter tuning. Multi-objective Bayesian optimization enables learn
 ing skill parameters that balance performance metrics such as safety\, spe
 ed\, and task success. Policies are trained in simulation and successfully
  transferred to real robots for contact-rich manipulation tasks.To support
  generalization\, the framework models task variations using gaussian proc
 esses\, enabling interpolation of BTMG parameters across unseen scenarios.
  This allows adaptive behavior without retraining for each new task instan
 ce.Failure recovery is addressed through a hierarchical scheme. BTs are ex
 tended with a reactive planner that dynamically updates execution policies
  based on runtime observations. Vision-language models assist in detecting
  and identifying failures\, and in generating symbolic corrections when ta
 sks are predicted to fail.The thesis concludes with a discussion of future
  work\, including (1) using vision-language-action (VLA) models or diffusi
 on policies to generate new skills on the fly from multimodal inputs\, and
  (2) extending the reactive planner with proactive failure prediction to a
 nticipate and prevent execution errors before they occur. Together\, these
  directions aim to advance robotic systems that are more robust\, adaptabl
 e\, and autonomous.&nbsp\;\n\nMer information om händelsen: https://www.c
 s.lth.se/evenemang/faseeh-ahmads-phd-defence
DTSTART;TZID=GMT:20251010T110000
DTEND;TZID=GMT:20251010T110000
LOCATION:E:1406\, E-huset\, Klas Anshelms väg 10/Ole Römers väg 3\, Lun
 d
END:VEVENT
END:VCALENDAR