OmniVLA Serbot: 82 Robot Navigation Episodes for Vision-Language-Action Learning

SERBOT navigation episodes collected for OmniVLA-style vision-language-action experiments. The dataset contains 82 valid episodes, 4,343 frames, and 4,342 actions across five distinct tasks. It was created by kjw6213 and last updated on June 4, 2026.

Use Cases

• Train vision-language-action models based on the described episodic navigation data.
• Benchmark robot navigation algorithms based on the five distinct tasks (task1 through task5).
• Develop imitation learning policies based on the 4,342 recorded actions.
• Study temporal action sequences based on the described action horizon of 8.

Strengths

• Contains 82 valid navigation episodes, providing a substantial number of training examples.
• Includes 4,342 actions with a defined 3-dimensional action space (linear_x, linear_y, angular_z).
• Data is organized into five distinct tasks (task1 through task5), enabling task-specific evaluation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

huggingface

Collection Method

Collected for OmniVLA-style vision-language-action experiments.

Freshness

Last updated 2026-06-04 07:18:52; freshness should be verified.