• Why?
  • In order to gain an understanding of more complex RL algorithms working with the FlexSim model, which includes state machine logic, multi-agent passenger management, and complex observation spaces
• How?
  • Read and understood how the learning algorithm interfaced with the FlexSim via the state machine diagrams
  • Read about how the system uses action masking (A vector to store decisions) to output decisions to act upon
  • Ran the corresponding Python files to see it in action (The environment, the training, and the inference based on the trained model)
  • Learned about command tensorboard --logdir=. in order to see the Tensorboard interface, and model results without using the command palette interface
  • Notice the difference in performance of the FlexSim model with the trained RL model making decisions rather than taking random decisions
• Importance?
  • Completing and understanding this allows for the bridging with previous more simpler concepts (From HelloWorld) to more complex agent models to understand in the future