Embodied AI Glossary中文RoboCasa
CommonA large-scale kitchen-household simulation framework and benchmark from UT Austin; its newest version is called RoboCasa365.
RoboCasa is a household-scene simulation framework from Yuke Zhu's group at UT Austin, built on top of robosuite and MuJoCo. Its first version, published at RSS 2024, included 120 kitchen scenes and 100 tasks, and used MimicGen (a tool that automatically expands a small number of human demonstrations into many more trajectories) to augment its data. RoboCasa365, released in February 2026 (ICLR 2026), expanded this to over 2,500 kitchen scenes, more than 3,200 objects, and 365 everyday tasks, with over 600 hours of human demonstrations and over 1,600 hours of automatically generated data, plus a multi-task leaderboard. It addresses the problem that real-robot household data is expensive and hard to reproduce, and is commonly used to compare generalist policies.
ExampleRoboCasa365's official leaderboard compares the multi-task performance of policies such as Diffusion Policy, the π series, and GR00T on the same set of kitchen tasks.
- Also called
- RoboCasa365
- Related
- robosuite · MimicGen · Simulation Assets · Benchmark · Household Tasks · NVIDIA Isaac GR00T N1
- Sources
- RoboCasa 官网（RoboCasa365） (Chinese)
RoboCasa: Large-Scale Simulation of Everyday Tasks for Generalist Robots (arXiv 2406.02523)
robocasa/robocasa GitHub README - As of
- 2026-05