VLM-Nav: UAV Navigation Performance Metrics from AirSim Simulation

Overall navigation performance of VLM-Nav, a vision-based UAV navigation method integrating depthmap estimation with Vision-Language Models. The dataset, authored by Gobinda Chandra Sarker and last updated in April 2026, contains simulation results from AirSim using Blocks and Downtown West environments. The system achieved a near-perfect task completion rate of 0.98.

Use Cases

• Benchmarking vision-language model performance for UAV navigation based on the described task completion rate
• Comparing obstacle avoidance strategies for autonomous vehicles based on the described depthmap and VLM integration
• Analyzing the cost-efficiency of sensor-less navigation systems based on the description of eliminating LiDAR
• Evaluating path planning algorithms in simulated complex environments like Blocks and Downtown West

Strengths

• Dataset is small (5.5 KB), facilitating quick download and inspection
• Specific performance metric reported: a task completion rate of 0.98
• Clear methodological description: integrates DepthAnything-V2 and VLMs like Gemini-flash/GPT-4o

Limitations

• Row count is unknown, which may limit suitability assessment
• Column-level documentation is absent; field semantics must be inferred after download
• Data is limited to simulation results from two specific AirSim environments

Provenance

Source

Gobinda Chandra Sarker via figshare

Collection Method

Generated from simulations in AirSim using Blocks and Downtown West environments

Time Range

Simulation date not specified; dataset updated April 2026

Freshness

Last updated 2026-04-03 17:54:47; freshness should be verified

Geography

Simulated environments (Blocks, Downtown West); no real-world geography