Out of Box Agent Algorithm Takes Dictionary Spaces

RLlib
1
  • High: It blocks me to complete my task.

Do we have an example of an out-of-box agent algorithm that handles dictionary spaces with various shapes?

An example of spaces using in environment:

action_bools = np.ones(self.n_jobs) + 1
self.action_space = gym.spaces.MultiDiscrete(action_bools)

# Observation space
self.observation_space = gym.spaces.Dict({
    'n_workers_avaliable': gym.spaces.Discrete(self.n_workers + 1),
    'action_mask': gym.spaces.Box(0, 1, shape=(self.n_jobs,), dtype=np.bool),
    'is_assigned': gym.spaces.Box(0, 1, shape=(self.n_jobs,), dtype=np.bool),
    'is_completed': gym.spaces.Box(0, 1, shape=(self.n_jobs,), dtype=np.bool),
    'job_obs': gym.spaces.Box(0.0, 1.0, shape=(self.n_jobs, 2), dtype=np.float32)
})

Thank you in advance!