Overview
What this challenge is about.
You receive an Isaac Sim navigation environment, a baseline trained policy, a 50-episode real-bench test set (recorded sensor streams + ground truth) for offline policy evaluation, and a budget of 8 training runs. Design a small factorial study: 4 randomization regimes (no randomization, lighting-only, full visual, full visual + dynamics) x 2 training durations. Evaluate each policy on (a) sim success rate (200 episodes), (b) real-bench success rate (offline scoring against the 50 held-out episodes), and (c) sim-to-real gap (sim - real). Recommend one regime for the next training cycle.
The Brief
What you'll do, and what you'll demonstrate.
Run a structured domain-randomization study and recommend the regime that minimizes sim-to-real gap for navigation policies.
Earning criteria — what you'll demonstrate
- Design a factorial domain-randomization study under a tight compute budget
- Evaluate sim-to-real gap correctly with a held-out real-bench set
- Reason about visual vs. dynamics randomization trade-offs
- Communicate sim-to-real findings to a robotics engineering team
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
Designing structured sim-to-real studies is the kind of research-engineering work robot-learning labs and startups hire for.
This challenge sharpens
- domain-randomization
- sim-to-real
- experiment-design
Research Scientist
Properly measuring sim-to-real gap with CIs and held-out benches is the rigor robotics research-scientist roles look for.
This challenge sharpens
- sim-to-real
- policy-evaluation
- experiment-design
Applied AI Scientist
Translating a randomization study into a recommended next training cycle is core applied-AI work at robotics startups.
This challenge sharpens
- domain-randomization
- robot-navigation
- policy-evaluation