todo: get foraging to work on argmax

[jarbus]

maybe add timestep input, so policies get something different each time?

• this is just the light level, no? edit: found light level was not included, I think

[jarbus]

policy does not change on different inputs for large models or reconstructed models with multiple seeds