Learning World Models from Egocentric Video for Home Robots
아직 한국어판이 없어 영어 원문으로 표시.
30초 요약
A video world model trained on robot egocentric footage predicts outcomes of actions for evaluation and planning.
연구 질문
Can world models replace physical evaluation?
문제
Evaluating home robots is slow and unsafe.
기존 접근
Physical rollouts or hand-built simulators.
새 접근
Action-conditioned video prediction.
결과
Correlates with real success rates on household tasks.
한계
Hallucinated physics; single robot.
업계 영향
Positions video generation vendors as robotics infrastructure.