WildOS: Visual Frontier Annotations for Outdoor Navigation

WildOS Frontiers contains between 1,000 and 10,000 images with visual frontier annotations for outdoor long-range navigation, released by leggedrobotics in February 2026. The dataset builds upon the GrandTour Dataset to identify exploration candidates such as trail ends, openings between trees, and road curves. It was developed to support open-vocabulary object search in unstructured wild environments.

Use Cases

• Training object detection models to identify trail ends and road curves as navigational targets
• Developing image segmentation algorithms to delineate openings between trees for autonomous path planning
• Benchmarking vision foundation models on traversability and frontier prediction in unstructured terrain

Strengths

• 1,000 to 10,000 annotated records
• Specialized frontier labels for long-range navigation
• Apache 2.0 permissive license

Limitations

• Geographic bias toward the United States as indicated by metadata tags
• Dependent on the environmental variety and sensor characteristics of the underlying GrandTour Dataset

Provenance

Source

leggedrobotics, associated with the WildOS research project

Collection Method

Expert-annotated on top of existing images from the GrandTour Dataset

Freshness

Last updated February 2026.

Geography

United States