AerialVLN-Fine Dataset

AerialVLN-Fine Overview

AerialVLN-Fine is a curated benchmark built from AerialVLN for more reliable zero-shot UAV VLN evaluation. It provides sentence-level alignment between instruction segments and trajectory segments, and refines ambiguous expressions with explicit visual endpoints and landmark references.

The dataset is designed to support fine-grained capability diagnosis and sentence-level evaluation, while maintaining high annotation quality through repeated manual verification.

AerialVLN-Fine is introduced as part of the paper FineCog-Nav: Integrating Fine-grained Cognitive Modules for Zero-shot Multimodal UAV Navigation.

Important Notice

This repository primarily provides the AerialVLN-Fine dataset files and format descriptions.

For:

• running agents / experiments
• evaluation scripts and protocols
• benchmark usage examples

please refer to the official FineCogNav repository:

https://github.com/SmartDianLab/FineCogNav

For additional details about the dataset, project background, examples, and other resources, please visit the project website:

https://smartdianlab.github.io/projects-FineCogNav/

Key Statistics

• Total trajectories: 300 high-quality instruction-trajectory pairs
• Source split: Val-Seen and Val-Unseen scenes from AerialVLN
• Average per trajectory: 189 meters and 76 actions
• Fine-grained semantics: 4.6 aligned instruction sentences per trajectory
• Total scale: 56,050 meters and 22,835 actions
• Instruction refinement: total words 17,572 to 30,762; average length 59 to 103
• Sentence-level total: 1,383 aligned sentences (average 41 meters, 17 actions)

Dataset Structure

Main dataset directory:

• AerialVLN-Fine/
  • scene_2/, scene_3/, scene_5/, ..., scene_24/
    • Sentence-level annotation JSON files (one trajectory per file)
  • TEST_FILE/
    • AerialVLN-format scene JSON files used for testing/evaluation

JSON Format 1: Scene Annotation Files

These files are under scene_xx folders, for example:

• AerialVLN-Fine-V3-20251110/scene_11/3180JW2OTDAQVE67WQ0FUSZAWP85JT.json

Each file contains one trajectory with sentence-level aligned segments.

Top-level fields

Key	Type	Description
episode_id	string	Unique episode identifier
trajectory_id	string	Unique trajectory identifier
scene_id	int	Scene index
instruction	string	Full refined instruction text
sentence_instructions	list	Sentence-level aligned annotation list
statistics	dict	Segment-level action statistics

statistics fields

Key	Type	Description
complete_actions	int	Total actions for this trajectory
sentence_actions	list[int]	Action counts of sentence segments
sentence_actions_average	float	Average actions per sentence segment

sentence_instructions item fields

Key	Type	Description
id	int	Sentence segment index
instruction	string	Original segment instruction text
completed_instruction	string	Refined/completed segment instruction text
start_frame	int	Start index in full trajectory (action/frame index)
end_frame	int	End index in full trajectory (action/frame index)
start_position	list[float]	Segment start position [x, y, z]
start_rotation	list[float]	Segment start quaternion [w, x, y, z]
end_position	list[float]	Segment end position [x, y, z]
end_rotation	list[float]	Segment end quaternion [w, x, y, z]
actions	list[int]	Action sequence for this sentence segment
reference_path	list[list[float]]	Segment trajectory points, each point is [x, y, z, roll, pitch, yaw]

JSON Format 2: AerialVLN-style TEST JSON (TEST_FILE)

These files are under TEST_FILE, for example:

• AerialVLN-Fine/TEST_FILE/AerialVLN-Fine-V3-Scene-11.json

Each file contains an episodes array in AerialVLN format.

Top-level fields

Key	Type	Description
episodes	list	List of AerialVLN episodes

episodes item fields

Key	Type	Description
episode_id	string	Unique episode identifier
trajectory_id	string	Unique trajectory identifier
scene_id	int	Scene index
start_position	list[float]	Start position [x, y, z]
start_rotation	list[float]	Start quaternion [w, x, y, z]
instruction	dict	Instruction container
goals	list[dict]	Goal list
reference_path	list[list[float]]	Full trajectory path points
actions	list[int]	Full action sequence

Nested fields in episodes

instruction:

• instruction_text: string

goals item:

• position: list[float] (goal position [x, y, z])

reference_path point:

• [x, y, z, roll, pitch, yaw]

Running and Evaluation

The execution pipeline, environment setup, and evaluation code are maintained in FineCogNav rather than this dataset repository.

Please use the following repository for running and evaluation:

https://github.com/SmartDianLab/FineCogNav

Please refer to the project website for additional details and updates:

https://smartdianlab.github.io/projects-FineCogNav/

Citation

If you use AerialVLN-Fine or FineCog-Nav in your research, please cite the corresponding project/paper information from the project website or the FineCogNav repository.

License

This dataset is licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0).

Acknowledgements

This dataset is built on top of AerialVLN and focuses on improving instruction clarity and sentence-level alignment quality for UAV VLN evaluation.

Some related code, evaluation utilities, and experimental pipelines are maintained in the FineCogNav project:

• https://github.com/SmartDianLab/FineCogNav

For more details, please visit:

• https://smartdianlab.github.io/projects-FineCogNav/