CognitiveDrone VLA Benchmark for UAV Trajectory Tasks

Over 8,000 simulated flight trajectories support a Vision-Language-Action model for Unmanned Aerial Vehicles. The dataset is structured for three key cognitive task categories: Human Recognition, Symbol Understanding, and Reasoning. It serves as an evaluation benchmark for real-time UAV task solving.

Use Cases

• Benchmark VLA model performance on Human Recognition tasks using simulated flight trajectories.
• Train models for Symbol Understanding by analyzing UAV trajectory patterns in synthetic environments.
• Evaluate reasoning capabilities for UAVs on tasks defined within the three core cognitive categories.
• Fine-tune models on the over 8,000 trajectory sequences to improve real-time decision-making.

Strengths

• Contains over 8,000 simulated flight trajectories for model training and evaluation.
• Focuses on three distinct, advanced cognitive task categories: Human Recognition, Symbol Understanding, and Reasoning.
• Designed specifically as a benchmark for Vision-Language-Action models in UAV applications.

Limitations

• Data is synthetic and simulated, which may not fully capture real-world flight conditions and noise.
• Specific column structure, data formats, and sample size per category are not detailed in the available input.

Provenance

Source

Hugging Face dataset authored by ArtemLykov.

Collection Method

Comprises simulated flight trajectories generated for model training.

Freshness

Last updated March 3, 2025.