Cross-Embodiment Robotics Dataset From 235 Community Contributors

791 distinct robotics datasets across 46 robot types provide a foundation for cross-embodiment pretraining of vision-language-action models. This crowdsourced collection was compiled by HuggingFaceVLA from 235 community contributors and was last updated in December 2025.

Use Cases

• Train vision-language-action models on multimodal data from 46 robot types to learn transferable manipulation skills.
• Benchmark cross-embodiment policy performance using the diverse task categories provided across the 791 constituent datasets.
• Pretrain models on the large-scale, community-sourced data to improve sample efficiency for downstream robotics tasks.

Strengths

• Compiles 791 distinct robotics datasets into a single resource.
• Encompasses data from 46 different robot types for broad embodiment coverage.
• Sourced from 235 community contributors, indicating a diverse data collection.

Limitations

• Specific dataset size, row counts, and file formats are unknown.
• The composition and balance of data across the 46 robot types is unspecified, risking representation bias.
• Lack of column information prevents assessment of feature consistency and data structure.

Provenance

Source

HuggingFaceVLA, building upon datasets used for SmolVLA.

Collection Method

Crowdsourced compilation from 235 open-source community contributors worldwide.

Freshness

Last updated December 10, 2025.