Imitation Learning from Human Demos for a Drone Inspection
Overview
What this challenge is about.
You receive 6 hours of expert pilot demonstrations (state-action pairs at 20 Hz) recorded in an AirSim wind-farm environment with 3 turbine designs, plus a held-out 4th turbine design for evaluation. Train a behavioral cloning (BC) policy as baseline, then a DAgger or Implicit Q-Learning policy. Evaluate on inspection completion rate, average distance from blade surface, and crash rate on the held-out turbine. Plot performance vs. dataset size (1h, 2h, 4h, 6h) and write a 2-page data-budget memo for the head of operations.
The Brief
What you'll do, and what you'll demonstrate.
Compare behavioral cloning vs. DAgger/IQL on expert demonstrations and quantify how much pilot data is needed for reliable autonomous inspection.
Earning criteria — what you'll demonstrate
- Implement behavioral cloning and a more advanced imitation-learning method
- Quantify generalization to held-out environments
- Diagnose distribution shift between expert and policy state distributions
- Translate a learning-curve into a data-collection budget
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
ML Researcher
Comparing imitation-learning methods and quantifying generalization to held-out environments is core ML research work for any autonomy team.
This challenge sharpens
- imitation-learning
- behavioral-cloning
- evaluation
Applied AI Scientist
Translating a learning curve into a procurement budget for expert pilot time is exactly what applied AI scientists do in robotics operations.
This challenge sharpens
- imitation-learning
- evaluation
- behavioral-cloning
Data Scientist
Building dataset-scaling curves and reporting them with confidence intervals to a non-ML operations lead transfers directly to data-science work.
This challenge sharpens
- evaluation
- dagger
- behavioral-cloning