Pre-lab

1. Begin formulating a strategy and training
   • Think of a strategy you want your robot to use to play the game
     • You may want to contact your group members and coordinate/discuss this (optional)
   • Begin experimenting with the strategy you chose (in the script that you renamed to your robot’s name from the last lab)
     • Add rewards for your robot as needed when implementing your strategy
   • (Optional) Check out this video which covers the ML methods we’ll cover in the lab that you can use to make your robot more effective.
   • Once the strategy is implemented, train your robot to obtain your first model
     • It is recommended that you train your robot for a while so meaningful behavior is established and can be evaluated
2. Submit the following for the pre-lab:
   • A brief description of the strategy you chose and how it was implemented (e.g. additional helpers you wrote, where rewards are added, etc.)
     • If you are unable to code all of it, that’s okay. Just say what you think will need to be implemented.
   • The model you obtained from the training (.nn OR .onnx)
   • How long it takes to train the model
   • Briefly evaluate the model you obtained. What does the robot seem to be doing well? What seems to be lacking? How can its performance be improved, strategy and implementation-wise?

Note that this is only your first attempt at training, and it is not expected to be perfect (or even functional). Just give it your best shot and see what can be improved on for the future.

Lab

• Your TA will address your questions from post-lab 7
• TA Demo: training settings
  1. Behavioural cloning
  2. Generative adversarial imitation learning (GAIL)
• TA Demo: modifying your robot appearance (by editing/adding prefabs)
• In your groups:
  1. Discuss each of your strategies and decide which of them you want to pursue further
  2. Adjust training settings: add behavioural cloning and/or GAIL and test out the training
     • Once you’re done: show your TA and explain what you did with the settings
  3. Tweak your strategy/implementation based on your analysis of your first model
  4. Modify your robot’s appearance and home base (by adjusting home base material)
  5. Show your TA your progress on one of 3 or 4

Postlab

1. What changes were made to strategy/implementation? Give a brief justification for your changes.
2. What training settings did you apply? Briefly explain what the settings does.