UnderspecifiedEnv

The UnderspecifiedEnv class defines a UPOMDP, and acts similarly to (but not identically to) a Gymnax environment.

The UnderspecifiedEnv class has the following interface

• params = env.default_params
• action_space = env.action_space(params)
• obs, state = env.reset_to_level(rng, level, params)
• obs, state, reward, done, info = env.step(rng, state, action, params)

Every environment must implement only the following methods

• step_env: Perform a step of the environment
• reset_env_to_level: Reset the environment to a particular level
• action_space: Return the action space of the environment

The environment also does not automatically reset to a new level once the environment has restarted. Look at the AutoReplay wrapper if this is desired.