EmbodiedNav-Bench: 5,037 Drone Navigation Trajectories in Urban 3D Airspace

EmbodiedNav-Bench is a goal-oriented embodied navigation benchmark containing 5,037 high-quality navigation trajectories. It includes natural-language navigation goals, initial drone poses, target positions, and ground-truth 3D trajectories for evaluation in urban 3D airspace. The dataset is hosted by EmbodiedCity on Hugging Face and was last updated in April 2026.

Use Cases

• Training embodied AI agents for goal-oriented navigation based on natural-language instructions.
• Benchmarking drone navigation algorithms using the provided 3D trajectories and target positions.
• Developing spatial reasoning models for urban airspace based on the initial drone poses and goals.
• Evaluating the performance of autonomous systems in simulated 3D environments.

Strengths

• Contains 5,037 high-quality navigation trajectories.
• Includes multiple aligned data modalities: natural-language goals, 3D poses, target positions, and ground-truth trajectories.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known, but other scale details like file size and formats are unknown.
• The description references an external project page for full details, requiring a click-out for complete context.

Provenance

Source

EmbodiedCity via Hugging Face.

Collection Method

Likely generated or collected within a simulation environment for benchmarking.

Freshness

Last updated 2026-04-25 02:55:51

Geography

Urban environments (implied by 'urban 3D airspace'); specific locations are unknown.