<aside> 👉 The Robustifying Autonomous Vehicles project aims to make autonomous vehicles safer and more robust by (1) designing interesting driving scenarios (2) implementing AI algorithms that can ensure safety and (3) explaining the behaviors post in training and post hoc.

</aside>

❄️ Winter/Spring 2024 Auditor Project

The Winter 2024 Auditor Project for the robustifying autonomous vehicles project will be to benchmark a suite of RL algorithms on an autonomous vehicle simulator of one’s choice.

• Week 1: Recruitment

Week 2: AIEA Lab Onboarding

Week 3: Onboarding + Nautilus

Week 4: Get an autonomous vehicle simulator running locally (Vista, Duckietown TORCS or CARLA).

• Carla is most the challenging, but the most adaptable to other problems.

Week 5: Get the simulator running on Nautilus in a headless environment (will need to be well acquainted on terminal). Document the process.

Week 6: Run a baseline RL algorithm from Stable Baselines and show the results in the simulated environment (note that the algorithm may need to be adapted for the specific simulation).

• MS students need to write an RL algorithm from scratch (of course, you can use Stable Baselines as an inspiration).

Week 7: Visualize the results from your RL algorithm run. What went well? What didn’t? You should have multiple graphs of the reward, etc.

Week 8: Write a generalized program to benchmark multiple RL algorithms in your simulation environment.

Week 9: Run the program for 2-3 RL algorithms, and report on the results.

Week 10: Refinement on the framework + report writing