[논문리뷰] WorldCycle: Self-Verifiable Reinforcement Learning for Long-Horizon Video World Models본 논문은 Interactive Video World Models (IWMs)이 long-horizon planning 및 exploration에서 겪는 compounding errors 문제를 해결하고자 합니다.#Review#Video World Models#Reinforcement Learning#Long-Horizon Planning#Compounding Errors#Self-Verification#Cycle-Consistency#Reversible Actions#Composite Actions#CycleBench2026년 8월 5일댓글 수 로딩 중