I'm referring to the environment @pengmun mentioned in #7.
A possible path to use simplified environments in our team approach would be not to train an RL agent there that hopefully generalises well to ksp (unlikely according to Ross), but generate a big dataset of evaluated runs that can be used to fine-tune a llm.
We need the environment to speak in the language of openai gym, same as the spacegym ones.
I'm referring to the environment @pengmun mentioned in #7.
A possible path to use simplified environments in our team approach would be not to train an RL agent there that hopefully generalises well to ksp (unlikely according to Ross), but generate a big dataset of evaluated runs that can be used to fine-tune a llm.
We need the environment to speak in the language of openai gym, same as the spacegym ones.