resources/tutorials/example for creating custom reward function?

[MicahJSherry]

I am having trouble translating my ideas for reward functions into bsk_rl objects. Can someone point me to a good (preferably beginner-friendly) example or tutorial that might help me adapt to my goals?

[Mark2000]

There aren't any tutorials specifically, but my advice would be to look at how one of the reward functions is implemented relative to the base data class. If you have a specific reward model in mind, I can perhaps offer some pointers on how I would structure it.

[MicahJSherry]

my goal is to have satellite that will match the orbit of a target satellite but I am unsure of how to implement it with the Data and Datastore classes.

[Mark2000]

Here's what I would do:

• Data: Contains the current difference in orbit from the desired orbit (in classical orbital elements, or Cartesian coordinates, or whatever). Addition would just produce a unit of data with the most recent measurement (may need to track the measurement time in the data to do this)
• DataStore: get_log_state would return the current OEs/coordinates. compare_log_states would only use the second argument (new_state) to generate an instance of your Data class that contains the difference between the state and the desired state.
• Rewarder: calculate_reward receives a dictionary with the satellite names and newest data, and needs to return a dictionary with satellite names and rewards.

Hope this helps. Check out some of the implemented data types for examples.

[MicahJSherry]

thanks for your help!