BridgeData2 LeRobot v3: Large-Scale Real-World Robot Manipulation Dataset

NVIDIA's BridgeData V2 is a large-scale real-world robotics dataset collected to support scalable robot learning. It has been converted into the LeRobotDataset v3.0 format. The dataset is designed for training policies conditioned on goal images or natural-language instructions.

Use Cases

• Train imitation learning models based on real-world robot manipulation demonstrations.
• Develop offline reinforcement learning agents based on a large-scale collection of robot interaction data.
• Build open-vocabulary multi-task robot policies based on goal image conditioning.
• Create natural-language-conditioned robot controllers based on the included instruction data.

Strengths

• Dataset is described as 'large-scale', suggesting substantial size.
• Specifically designed for multiple robot learning paradigms, including imitation learning and offline RL.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last updated 2026-05-31 19:04:48; freshness should be verified.

Provenance

Source

NVIDIA

Collection Method

Collected from real-world robot manipulation tasks.

Freshness

Last updated 2026-05-31 19:04:48.