Train a Reinforcement-Learning Locomotion Policy for a Quadruped
Overview
What this challenge is about.
You receive a configured Isaac Lab environment for the quadruped, a baseline PPO trainer, and a set of 8 trip-hazard / slip stress scenarios. Train the policy for a budget of about 200 million sim steps. Evaluate success rate, average forward velocity, and joint torques on each stress scenario. Document the reward-shaping choices you made and the failure modes that remain. Deliver the trained policy checkpoint, evaluation report, and a 3-page research write-up.
The Brief
What you'll do, and what you'll demonstrate.
Train an RL locomotion policy that crosses 5cm trip hazards and recovers from slips with high success across a stress-test suite.
Earning criteria — what you'll demonstrate
- Train an RL locomotion policy with PPO and domain randomization
- Design a stress-test suite that captures real deployment hazards
- Document reward-shaping choices traceable to deployment outcomes
- Communicate research results to engineering leadership
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
RL locomotion with rigorous stress evaluation and reward-shaping documentation is the daily reality of an applied ML researcher in legged-robot teams.
This challenge sharpens
- reinforcement-learning
- locomotion
- policy-evaluation
Research Scientist
Ablation-driven analysis of domain randomization channels mirrors the standards of a junior research-scientist's first publishable project.
This challenge sharpens
- reinforcement-learning
- domain-randomization
- policy-evaluation
Applied AI Scientist
Bridging from research-grade RL training to a deployment-grade write-up is the applied-AI scientist's core craft on robotics teams.
This challenge sharpens
- reinforcement-learning
- domain-randomization
- locomotion