[rllib] observation function api for multi-agent (#8236)

doc/source/rllib-env.rst

@@ -253,9 +253,9 @@ The most general way of implementing a centralized critic involves modifying the
 
 To update the critic, you'll also have to modify the loss of the policy. For an end-to-end runnable example, see `examples/centralized_critic.py <https://github.com/ray-project/ray/blob/master/rllib/examples/centralized_critic.py>`__.
 
-**Strategy 2: Sharing observations through the environment**:
+**Strategy 2: Sharing observations through an observation function**:
 
-Alternatively, the env itself can be modified to share observations between agents. In this strategy, each observation includes all global state, and policies use a custom model to ignore state they aren't supposed to "see" when computing actions. The advantage of this approach is that it's very simple and you don't have to change the algorithm at all -- just use an env wrapper and custom model. However, it is a bit less principled in that you have to change the agent observation spaces and the environment. You can find a runnable example of this strategy at `examples/centralized_critic_2.py <https://github.com/ray-project/ray/blob/master/rllib/examples/centralized_critic_2.py>`__.
+Alternatively, you can use an observation function to share observations between agents. In this strategy, each observation includes all global state, and policies use a custom model to ignore state they aren't supposed to "see" when computing actions. The advantage of this approach is that it's very simple and you don't have to change the algorithm at all -- just use the observation func (i.e., like an env wrapper) and custom model. However, it is a bit less principled in that you have to change the agent observation spaces to include training-time only information. You can find a runnable example of this strategy at `examples/centralized_critic_2.py <https://github.com/ray-project/ray/blob/master/rllib/examples/centralized_critic_2.py>`__.
 
 Grouping Agents
 ~~~~~~~~~~~~~~~