RL: Extensions on Double Pendulum

[detlefarend]

As discussed in our today's meeting:

1. Extensions on rewarding

• [x] New parameter p_rst_balancing
• [x] New parameter p_rst_swinging
• [x] Constants for valid reward strategies for balancing
• [x] Constants for valid reward strategies for swinging
• [x] Internal dictionary to store reward_methods for state area (balancing/swinging)
• [x] New reward strategy for balancing including old/new state

2. Extensions on start position

• [x] New optional parameter for boundaries of random start position
• [x] New optional parameter for boundaries of balancing state area

3. Further extensions

• [x] New boolean parameter: break when leaving the balancing state area
• [x] New optional horizon parameter for reward plot (if > 0 then this number of cycles is shown in the plot
• [x] New optional boolean parameter for an additional trend line
• [x] New optional parameter for state component weights (Postponed!)

4. Documentation

• [x] Howto1 with random agent
• [x] Howto2 for manual env validation
• [x] RTD: Description of env and howtos

[detlefarend]

Hi @laxmikantbaheti, it seems that this issue is done so far. Can we close it?

[laxmikantbaheti]

Hi @detlefarend yes the marked tasks are done. The weighted state space is yet to be implemented.

It's also true that this wouldn't take a lot of time. I just lost it in other issues. Shall I quickly finish this and close the issue?

[detlefarend]

Hi @laxmikantbahet, sure. Would be good👍

[laxmikantbaheti]

Done.