Skip to main content
Datasetroboticsv1.0

CALVIN

by Albert-Ludwigs-Universität Freiburg · open-source · Last verified 2026-03-17

CALVIN (Composing Actions from Language and Vision) is a dataset and benchmark for training and evaluating language-conditioned, long-horizon robot manipulation policies. It provides 24 hours of play-data demonstrations and 34 manipulation skills in a tabletop environment, with natural language instructions that challenge models to chain multiple subtasks from open-ended commands.

https://calvindataset.github.io/CALVIN
C+
C+Average
Adoption: BQuality: AFreshness: BCitations: B+Engagement: F

Specifications

License
MIT
Pricing
open-source
Capabilities
language-conditioned-manipulation, long-horizon-planning, play-data-learning
Integrations
Use Cases
language-robot-policy-training, long-horizon-task-chaining, embodied-instruction-following
API Available
No
Tags
robotics, language-conditioned, manipulation, long-horizon, benchmark
Added
2026-03-17
Completeness
100%

Index Score

59.5
Adoption
60
Quality
84
Freshness
68
Citations
75
Engagement
0

Put AI to work for your business

Deploy this dataset alongside autonomous AaaS agents that handle tasks end-to-end — no babysitting required.

Explore the full AI ecosystem on Agents as a Service