VLM-Nav Inference Time Data for UAV Navigation Simulations

A 5.5 KB dataset by Gobinda Chandra Sarker, last updated in April 2026, containing inference time measurements for a vision-based UAV navigation system. The system integrates depthmap estimation from RGB images with Vision-Language Models for obstacle avoidance and path planning, validated in AirSim simulations.

Use Cases

• Benchmarking inference speed of Vision-Language Models (Gemini-flash, GPT-4o) for real-time navigation tasks based on the described system.
• Analyzing the performance of depthmap estimation (DepthAnything-V2) integrated with VLMs for obstacle avoidance based on the study's methodology.
• Comparing navigation system efficiency in different simulated environments (Blocks, Downtown West) as mentioned in the validation.
• Evaluating cost-efficient navigation solutions that operate without LiDAR or pre-existing maps, a key feature described.

Strengths

• Dataset is openly licensed under CC-BY-4.0.
• Specific simulation environments (AirSim Blocks and Downtown West) and a high task completion rate (0.98) are documented.
• File size is precisely known (5.5 KB).

Limitations

• Row count and column definitions are unknown, limiting suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Data is very small in scale (5.5 KB), indicating limited scope.

Provenance

Source

figshare, author Gobinda Chandra Sarker.

Collection Method

Likely generated from simulations of a proposed UAV navigation system.

Time Range

null

Freshness

Last updated 2026-04-03 17:54:48; freshness should be verified.

Geography

null