Python: Agent.act()

[mensch72]

provide act() and observe() methods in Agent that can be called by the simulator in this fashion, so that the simulator does not need to care about aspirations at all:

state, _ = env.reset()
agent.observe(state)
while not terminated:
    action = agent.act() 
    state, delta, terminated, _, _ = env.step(action)
    agent.observe(state, delta, terminated)

Note: act() can later also contain code for consulting with the principal before returning an action. observe() might later also contain code for updating a learned world model.