Skip to content
All projects

Robotics · 2024 to present · Research

AeroVLA

A vision-language mission planner that turns natural-language requests into drone flight paths, planning 6.5x faster than a manual workflow.

Overview

AeroVLA closes the gap between human intent and drone mission setup. It uses vision-language reasoning and task prompts to generate workable mission paths from plain-text commands.

My role

System design, dataset work, the prompt and planning pipeline, and mission workflow design.

Highlights

  • Cut planning time by 6.5x against a manual workflow
  • Improved trajectory efficiency by 22 percent in evaluation runs
  • Built the UAV-VLPA-nano-30 dataset to support the planner