RoboCasa365: A Large-Scale Simulation Framework for Training and Benchmarking Generalist Robots

Nasiriany, Nasiriany, Maddukuri, Zhu · 2026 · arXiv preprint · arXiv:2603.04356 · PDF

Dongyu supplement. Added as a candidate missing/adjacent dataset paper after searching primary arXiv sources. This is a draft summary for triage, not a full paper read.

One-liner. A 2026 RoboCasa extension with 365 household mobile-manipulation tasks, 2,500 kitchen environments, and large human/synthetic demonstration corpora.

Setup

• Datasets / benchmarks: Introduces RoboCasa365: 365 everyday tasks across 2,500 kitchen environments, with over 600 hours of human demonstration data and over 1,600 hours of synthetic demonstration data.
• Hardware / simulator: Simulation benchmark for household mobile manipulation, built on RoboCasa.

Method

Large-scale benchmark for multi-task learning, robot foundation model training, and lifelong learning.

Why it matters for the map

Good latest benchmark reference for scale and systematic evaluation.

Limitations / open questions

Still simulation-centric and not focused on language-grounded non-visual dynamic concepts.

Source note

arXiv search result reports 365 tasks, 2,500 environments, and demonstration-hour scale.