Robot Arm Teleoperation Data for Vision-Language-Action Finetuning

A demonstration dataset for Vision-Language-Action model finetuning, collected via teleoperation of a robot arm performing a basic pick-and-place task. The data was gathered by author IliaLarchenko using a modified LeKiwi setup with three cameras and was last updated in September 2025.

Use Cases

• Fine-tuning a VLA model to predict robot actions from video frames of the arm and scene.
• Training an imitation learning policy using teleoperated video demonstrations of pick-and-place tasks.
• Benchmarking robot manipulation models on data collected with the LeKiwi system and SO-ARM100 configuration.

Strengths

• Data collected using a 3-camera setup for multi-view recording.
• Specifically designed for the SO-ARM100 robot arm configuration.

Limitations

• Dataset size and row count are unknown, potentially limiting model training.
• Contains only a single basic pick-and-place task, reducing task diversity.

Provenance

Source

IliaLarchenko via Hugging Face.

Collection Method

Collected via human teleoperation of a robot arm using a modified LeKiwi system with 3 cameras.

Freshness

Last updated September 2025.

Geography

Region tags suggest 'us', but specific location is unknown.