RecreationWorld: 하이브리드 컴퓨터 사용 에이전트를 위한 확장 가능하고 검증 가능한 환경
이 논문은 GUI 상호작용과 코드 기반 소프트웨어 개발을 동시에 수행하는 hybrid CUAs를 연구합니다. 이를 위해 reference의 동작을 발견하고 충실한 구현체를 구축하는 five-platform framework인 RecreationWorld를 제안합니다. 이 프레임워크는 Ubuntu, macOS, Windows, Android, Web 환경에서 reproducible environments와 unified harness를 제공하여 모델 학습을 지원합니다.
Computer-use agents (CUAs) have advanced along two separate lines: graphical interaction and software development through code and the command line. Real digital work requires both, interleaved rather than stacked end to end. We study hybrid CUAs that autonomously decide when to explore an interface, implement software, and run and visually verify their artifacts. We introduce RecreationWorld, a five-platform framework built around recreation: given a running reference, an agent must discover its behavior and build a faithful implementation with no prescribed workflow. RecreationWorld provides reproducible environments on Ubuntu, macOS, Windows, Android, and Web, plus a unified harness with native GUI control and coding tools. The running reference serves as an oracle for hidden behavioral tests, providing execution-grounded rewards. We scale trajectory generation with high-quality open-source applications. Models trained on these trajectories improve across five out-of-distribution co