Imitation Learning from Human Demos for a Drone Inspection
Overview
What this challenge is about.
Train BC and DAgger policies on drone inspection demos, then write a data-budget memo. Get a verifiable certificate.
The scenario
The startup (around 25 people, 90 wind farms served across France, Spain, and Portugal) currently collects expert demos at roughly EUR 600 per pilot-hour; knowing where the data-vs-performance curve flattens directly sizes the data-collection budget.
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.
Deep Reinforcement Learning
Master · Reinforcement Learning
Strong alignment
This challenge maps to Deep Reinforcement Learning at the Master level. It sharpens the same practical skills your coursework expects — but in a real industry context with actual constraints and deliverables.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
- Imitation Learning
Apply imitation learning to solve real industry problems and demonstrate production-level capability.
- Behavioral Cloning
Apply behavioral cloning to solve real industry problems and demonstrate production-level capability.
- Dagger
Apply dagger to solve real industry problems and demonstrate production-level capability.
- Reinforcement Learning
Apply reinforcement learning to solve real industry problems and demonstrate production-level capability.
- Pytorch
Apply pytorch to solve real industry problems and demonstrate production-level capability.
- Evaluation
Apply evaluation to solve real industry problems and demonstrate production-level capability.
Careers
Career paths this challenge builds toward
Completing this challenge demonstrates skills that transfer directly to these roles:
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