Persistent data and time in reward_function

Question

I want to identify events such as the car changing direction or crossing a threshold and calculate how frequently those events happen.  This will require a history of values and some knowledge of time.  Is there any way to get this data within the reward function?  From the examples it seems like we only have the ability to know the cars instantaneous state.

Answer

I suspected that could be the case, will give it a shot.  Thanks!

Answer

This is not necessarily the answer you are looking for, but you can do this if you use the SageMaker Jupyter notebook method to train your model.   In fact the code already does this, so you can just follow the patterns in there.  
  
From what I can see this will not work in the DeepRacer console, but I have not tried very hard.

Persistent data and time in reward_function

Conteúdo relevante