Virtual Qualifier 1 and 2

Anduril runs its drone racing qualifiers inside a simulator built by Drone Champions League. I flew both virtual rounds with a Python autonomy stack talking MAVLink. The whole job is to see the gate, pass it cleanly, and never invent geometry that is not there.

VQ1 is the on ramp. Fewer gates, a flatter looking world, gates that are visually highlighted, and an eight minute cap. Completion is what counts. I used that round to lock down perception and control with OpenCV and a YOLO gate keypoint pipeline fed by Label Studio.

VQ2 removes the help. Harder lighting, a real 3D scanned environment, more gates, and time as the ranking. I moved off the VQ1 approach of a hardcoded world path and went to body space gate tracking, so the same policy could survive a course it had never memorized.

I did not qualify for physical qualifier 1. Reinforcement learning and better state estimation are what I want to get right before I try again.