EmbodiedHubEmbodied AI Datasets
A collection of key datasets used for training embodied AI models, including robot manipulation and vision-language-action models.
Ecosystem Snapshot
Leading Datasets
BridgeData V2 is a large-scale robot manipulation dataset from UC Berkeley RAIL Lab with 60,000+ WidowX trajectories across diverse scenes for training generalist robot policies.
Largest high-resolution dataset of real-world 3D indoor environments with 1,000 spaces covering 215 million m², designed as the backbone for embodied AI navigation and interaction research.
Largest publicly available dataset of human demonstrations in Minecraft with 60+ million action frames, designed for sample-efficient imitation and reinforcement learning.
Universal dexterous grasping dataset with 100K+ grasp proposals across 2,846 articulated objects, using a diffusion-based generative model for diverse grasp generation.
Platform and dataset for dexterous manipulation using synchronized human video demonstrations and simulated Shadow Hand trajectories with 50+ object manipulation tasks.
First comprehensive benchmark for whole-body humanoid control with 30 standardized evaluation tasks combining locomotion and manipulation in MuJoCo simulation.
Leading Benchmarks
Industry Insights
This page aggregates the core open datasets that underpin embodied AI research and VLA model training. These datasets provide diverse, large-scale robot demonstration data spanning multiple robot embodiments, tasks, scenes, and environments.
Key datasets include Open X-Embodiment (1M+ episodes across 22 robot types), DROID (87K+ in-the-wild trajectories), and BridgeData (60K+ WidowX demonstrations for generalization research).