RLlib works with several different types of environments, including `OpenAI Gym <https://gym.openai.com/>`__, user-defined, multi-agent, and also batched environments.
In the high-level agent APIs, environments are identified with string names. By default, the string will be interpreted as a gym `environment name <https://gym.openai.com/envs>`__, however you can also register custom environments by name:
In the above example, note that the ``env_creator`` function takes in an ``env_config`` object. This is a dict containing options passed in through your agent. You can also access ``env_config.worker_index`` and ``env_config.vector_index`` to get the worker id and env id within the worker (if ``num_envs_per_worker > 0``). This can be useful if you want to train over an ensemble of different environments, for example:
RLlib uses Gym as its environment interface for single-agent training. For more information on how to implement a custom Gym environment, see the `gym.Env class definition <https://github.com/openai/gym/blob/master/gym/core.py>`__. You may also find the `SimpleCorridor <https://github.com/ray-project/ray/blob/master/python/ray/rllib/examples/custom_env.py>`__ and `Carla simulator <https://github.com/ray-project/ray/blob/master/python/ray/rllib/examples/carla/env.py>`__ example env implementations useful as a reference.
1.**Vectorization within a single process:** Though many envs can achieve high frame rates per core, their throughput is limited in practice by policy evaluation between steps. For example, even small TensorFlow models incur a couple milliseconds of latency to evaluate. This can be worked around by creating multiple envs per process and batching policy evaluations across these envs.
You can configure ``{"num_envs_per_worker": M}`` to have RLlib create ``M`` concurrent environments per worker. RLlib auto-vectorizes Gym environments via `VectorEnv.wrap() <https://github.com/ray-project/ray/blob/master/python/ray/rllib/env/vector_env.py>`__.
2.**Distribute across multiple processes:** You can also have RLlib create multiple processes (Ray actors) for experience collection. In most algorithms this can be controlled by setting the ``{"num_workers": N}`` config.
You can also combine vectorization and distributed execution, as shown in the above figure. Here we plot just the throughput of RLlib policy evaluation from 1 to 128 CPUs. PongNoFrameskip-v4 on GPU scales from 2.4k to ∼200k actions/s, and Pendulum-v0 on CPU from 15k to 1.5M actions/s. One machine was used for 1-16 workers, and a Ray cluster of four machines for 32-128 workers. Each worker was configured with ``num_envs_per_worker=64``.
RLlib will auto-vectorize Gym envs for batch evaluation if the ``num_envs_per_worker`` config is set, or you can define a custom environment class that subclasses `VectorEnv <https://github.com/ray-project/ray/blob/master/python/ray/rllib/env/vector_env.py>`__ to implement ``vector_step()`` and ``vector_reset()``.
A multi-agent environment is one which has multiple acting entities per step, e.g., in a traffic simulation, there may be multiple "car" and "traffic light" agents in the environment. The model for multi-agent in RLlib as follows: (1) as a user you define the number of policies available up front, and (2) a function that maps agent ids to policy ids. This is summarized by the below figure:
..image:: multi-agent.svg
The environment itself must subclass the `MultiAgentEnv <https://github.com/ray-project/ray/blob/master/python/ray/rllib/env/multi_agent_env.py>`__ interface, which can returns observations and rewards from multiple ready agents per step:
"traffic_light" # Traffic lights are always controlled by this policy
if agent_id.startswith("traffic_light_")
else random.choice(["car1", "car2"]) # Randomly choose from car policies
},
},
})
while True:
print(trainer.train())
RLlib will create three distinct policies and route agent decisions to its bound policy. When an agent first appears in the env, ``policy_mapping_fn`` will be called to determine which policy it is bound to. RLlib reports separate training statistics for each policy in the return from ``train()``, along with the combined reward.
Here is a simple `example training script <https://github.com/ray-project/ray/blob/master/python/ray/rllib/examples/multiagent_cartpole.py>`__ in which you can vary the number of agents and policies in the environment. For how to use multiple training methods at once (here DQN and PPO), see the `two-trainer example <https://github.com/ray-project/ray/blob/master/python/ray/rllib/examples/multiagent_two_trainers.py>`__. Metrics are reported for each policy separately, for example:
To scale to hundreds of agents, MultiAgentEnv batches policy evaluations across multiple agents internally. It can also be auto-vectorized by setting ``num_envs_per_worker > 1``.
RLlib will create each policy's model in a separate ``tf.variable_scope``. However, variables can still be shared between policies by explicitly entering a globally shared variable scope with ``tf.VariableScope(reuse=tf.AUTO_REUSE)``:
There is a full example of this in the `example training script <https://github.com/ray-project/ray/blob/master/python/ray/rllib/examples/multiagent_cartpole.py>`__.
Implementing a Centralized Critic
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Implementing a shared critic between multiple policies requires the definition of custom policy graphs. It can be done as follows:
1. Querying the critic: this can be done in the ``postprocess_trajectory`` method of a custom policy graph, which has full access to the policies and observations of concurrent agents via the ``other_agent_batches`` and ``episode`` arguments. This assumes you use variable sharing to access the critic network from multiple policies. The critic predictions can then be added to the postprocessed trajectory. Here's an example:
# metrics like "global reward" can be retrieved from the info return of the environment
sample_batch["global_rewards"] = [
info["global_reward"] for info in sample_batch["infos"]]
return sample_batch
2. Updating the critic: the centralized critic loss can be added to the loss of some arbitrary policy graph. The policy graph that is chosen must add the inputs for the critic loss to its postprocessed trajectory batches.
For an example of defining loss inputs, see the `PGPolicyGraph example <https://github.com/ray-project/ray/blob/master/python/ray/rllib/agents/pg/pg_policy_graph.py>`__.
In many situations, it does not make sense for an environment to be "stepped" by RLlib. For example, if a policy is to be used in a web serving system, then it is more natural for an agent to query a service that serves policy decisions, and for that service to learn from experience over time.
RLlib provides the `ServingEnv <https://github.com/ray-project/ray/blob/master/python/ray/rllib/env/serving_env.py>`__ class for this purpose. Unlike other envs, ServingEnv has its own thread of control. At any point, agents on that thread can query the current policy for decisions via ``self.get_action()`` and reports rewards via ``self.log_returns()``. This can be done for multiple concurrent episodes as well.
For example, ServingEnv can be used to implement a simple REST policy `server <https://github.com/ray-project/ray/tree/master/python/ray/rllib/examples/serving>`__ that learns over time using RLlib. In this example RLlib runs with ``num_workers=0`` to avoid port allocation issues, but in principle this could be scaled by increasing ``num_workers``.
Offline Data
~~~~~~~~~~~~
ServingEnv also provides a ``self.log_action()`` call to support off-policy actions. This allows the client to make independent decisions, e.g., to compare two different policies, and for RLlib to still learn from those off-policy actions. Note that this requires the algorithm used to support learning from off-policy decisions (e.g., DQN).
The ``log_action`` API of ServingEnv can be used to ingest data from offline logs. The pattern would be as follows: First, some policy is followed to produce experience data which is stored in some offline storage system. Then, RLlib creates a number of workers that use a ServingEnv to read the logs in parallel and ingest the experiences. After a round of training completes, the new policy can be deployed to collect more experiences.
Note that envs can read from different partitions of the logs based on the ``worker_index`` attribute of the `env context <https://github.com/ray-project/ray/blob/master/python/ray/rllib/env/env_context.py>`__ passed into the environment constructor.
Batch Asynchronous
------------------
The lowest-level "catch-all" environment supported by RLlib is `AsyncVectorEnv <https://github.com/ray-project/ray/blob/master/python/ray/rllib/env/async_vector_env.py>`__. AsyncVectorEnv models multiple agents executing asynchronously in multiple environments. A call to ``poll()`` returns observations from ready agents keyed by their environment and agent ids, and actions for those agents can be sent back via ``send_actions()``. This interface can be subclassed directly to support batched simulators such as `ELF <https://github.com/facebookresearch/ELF>`__.
Under the hood, all other envs are converted to AsyncVectorEnv by RLlib so that there is a common internal path for policy evaluation.
