BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Continual Learning for Embodied Agents in Open-Ended Environments
DTSTART:20260204T140000
DTEND:20260204T160000
DTSTAMP:20260403T235722Z
UID:3d69c5d4765d7497edd0c826aba446a0abfc9153d0e0f4dc9216ff4d
CATEGORIES:Conferences - Seminars
DESCRIPTION:Chengkun Li\nEDIC candidacy exam\nExam president: Prof. Wulfra
 m Gerstner\nThesis advisor: Prof. Alexander Mathis\nCo-examiner: Prof. Auk
 e Ijspeert\n\nAbstract\nEmbodied agents deployed in the real world face bi
 g worlds: open-ended\, non-stationary environments whose task distribution
 s cannot be fully captured by any fixed benchmark. Yet most reinforcement 
 learning studies still rely on static task suites\, which makes it difficu
 lt to evaluate meaningful progress when goals\, skills\, and environments 
 evolve over time.\n\nWe hypothesize that progress in open-ended embodied i
 ntelligence must start with evaluation: principled ways to generate\, sele
 ct\, and assess tasks that reveal what an agent can reliably do\, what it 
 forgets\, and where it fails. This PhD\, Continual Learning for Embodied A
 gents in Open-Ended Environments\, will build on model-driven open-endedne
 ss\, where foundation models propose novel tasks and produce executable en
 vironment specifications guided by notions of novelty and interestingness.
  We will develop evaluation protocols\, metrics\, and diagnostic tests for
  open-ended progress\, including robustness under distribution shift and s
 ensitivity to reward or success-specification errors.\n\nRecent advances i
 n scalable simulated control platforms\, including modern MuJoCo-based sui
 tes\, enable systematic long-horizon experimentation. Using these tools\, 
 we will iteratively refine the evaluation framework and\, guided by its fi
 ndings\, progressively develop methods that improve stability and adaptati
 on in big worlds.\n\nBackground papers\n\n	The Big World Hypothesis and it
 s Ramifications for Artificial Intelligence\n	https://openreview.net/forum
 ?id=Sv7DazuCn8\n	OMNI-EPIC: Open-endedness via Models of Human Notions of 
 Interestingness with Environments Programmed in Code\n	https://openreview.
 net/forum?id=Y1XkzMJpPd\n	MuJoCo Playground\n	https://arxiv.org/abs/2502.0
 8844\n
LOCATION:
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR