Grid2Op
Grid2Op copied to clipboard
Something that would "do nothing" if the agent took too long to take a decision
Is your feature request related to a problem? Please describe.
This is currently not possible to do.
The idea of the feature is that we could pass an option to the runner (or even better a different Runner class) that would "do nothing" if the agent takes too long to decide what to do.
Basically, say we impose a decision time of 0.1s / step:
- if an agent takes less than 0.1 s to send its action then the environment runs normally
- if an agent takes between 0.10000000001 and 0.2s then the environment does nothing for one step, then does the action of the agent, then sends the observation to the agent
- f an agent takes between 0.20000000001 and 0.3s then the env does nothing for 2 steps, the does the action, then sends the observation (after the 2 do nothing)
A specific environment class that does that (without "modifying" the Runner would also be a great addition).
Interest
This feature would be closer to what happens in real time: if you take too long to take your decision "time goes on" and it's like you do nothing for a steps.
This would be a nice addition for future L2RPN competition as well.