RoboCasa is a large-scale simulation framework for training generalist robots in everyday environments. It features realistic and diverse kitchen scenes, thousands of 3D assets, and cross-embodiment support for mobile manipulators and humanoid robots. The framework includes 100 tasks, including 25 atomic tasks for foundational skills and 75 composite tasks generated using large language models. RoboCasa uses generative AI tools to create environment textures and 3D objects, and provides high-quality human demonstrations and automated trajectory generation methods to expand datasets. The framework supports a wide range of kitchen activities, including cooking, cleaning, and restocking. Experiments show that synthetically generated data can significantly improve robot policy learning and transfer to real-world tasks. RoboCasa also includes a large multi-task dataset of 100K+ trajectories, enabling systematic evaluation of robot performance. The framework is designed to support a variety of robot embodiments, including mobile manipulators, humanoid robots, and quadruped robots with arms. RoboCasa is built on RoboSuite and extends its capabilities with a large array of scenes, objects, and hardware platforms. The framework includes a diverse set of kitchen styles and floor plans, and supports a wide range of kitchen appliances and objects. The simulation is designed to be realistic and diverse, with photorealistic rendering and a large collection of tasks. The framework is intended to facilitate rapid prototyping and reproducible research in robot learning. RoboCasa is a promising tool for training generalist robots in everyday environments, with the potential to significantly improve robot learning and deployment in real-world tasks.RoboCasa is a large-scale simulation framework for training generalist robots in everyday environments. It features realistic and diverse kitchen scenes, thousands of 3D assets, and cross-embodiment support for mobile manipulators and humanoid robots. The framework includes 100 tasks, including 25 atomic tasks for foundational skills and 75 composite tasks generated using large language models. RoboCasa uses generative AI tools to create environment textures and 3D objects, and provides high-quality human demonstrations and automated trajectory generation methods to expand datasets. The framework supports a wide range of kitchen activities, including cooking, cleaning, and restocking. Experiments show that synthetically generated data can significantly improve robot policy learning and transfer to real-world tasks. RoboCasa also includes a large multi-task dataset of 100K+ trajectories, enabling systematic evaluation of robot performance. The framework is designed to support a variety of robot embodiments, including mobile manipulators, humanoid robots, and quadruped robots with arms. RoboCasa is built on RoboSuite and extends its capabilities with a large array of scenes, objects, and hardware platforms. The framework includes a diverse set of kitchen styles and floor plans, and supports a wide range of kitchen appliances and objects. The simulation is designed to be realistic and diverse, with photorealistic rendering and a large collection of tasks. The framework is intended to facilitate rapid prototyping and reproducible research in robot learning. RoboCasa is a promising tool for training generalist robots in everyday environments, with the potential to significantly improve robot learning and deployment in real-world tasks.