Update docs

Author: ondrej-lukas

docs/Coordinator.md

@@ -14,13 +14,17 @@ Coordinator, having the role of the middle man in all communication between the
 
 1. `Actions queue` is a queue in which the agents submit their actions. It provides N:1 communication channel in which the coordinator receives the inputs.
 2. `Answer queue` is a separeate queue **per agent** in which the results of the actions are send to the agent.
-3.  
+3.  `World action queue` is a queue used for sending the acions from coordinator to the AI Dojo world
+4. `World response queue` is a channel used for wolrd -> coordinator communicaiton (responses to the agents' action)
 <img src="/docs/figures/message_passing_coordinator.jpg" alt="Message passing overview" width="30%"/>
 
 
 ## Main components of the coordinator
-`self._actions_queue`: asycnio queue for agent -> aidojo_world communication
-`self._answers_queue`: asycnio queue for aidojo_world -> agent communication
+`self._actions_queue`: asycnio queue for agents -> coordinator communication
+`self._answer_queues`: dictionary of asycnio queues for coordinator -> agent communication (1 queue per agent)
+`self._world_action_queue`: asycnio queue for coordinator -> world  queue communication
+`self._world_response_queue`: asycnio queue for world -> coordinator  queue communication
+`self.task_config`: Object with the configuration of the scenario
 `self.ALLOWED_ROLES`: list of allowed agent roles [`Attacker`, `Defender`, `Benign`]
 `self._world`: Instance of `AIDojoWorld`. Implements the dynamics of the world   
 `self._CONFIG_FILE_HASH`: hash of the configuration file used in the interaction (scenario, topology, etc.). Used for better reproducibility of results
@@ -33,33 +37,11 @@ Coordinator, having the role of the middle man in all communication between the
 ### Agent information components
 `self.agents`: information about connected agents {`agent address`: (`agent_name`,`agent_role`)}
 `self._agent_steps`: step counter for each agent in the current episode
-`self._reset_requests`: dictionary where requests for episode reset are collected (the world resets only if ALL agents request reset)
+`self._reset_requests`: dictionary where requests for episode reset are collected (the world resets only if **all** active agents request reset)
 `self._agent_observations`: current observation per agent
 `self._agent_starting_position`: starting position (with wildcards, see [configuration](../README.md#task-configuration)) per agent
 `self._agent_states`: current GameState per agent 
-`self._agent_statuses`: status of each agent. One of following options:
-    - `playing`: agent is registered and can participate in current episode. Can't influence the episode termination
-    - `playing_active`: agent is registered and can participate in current episode. It has `goal` and `max_steps` defined and can influence the termination of the episode
-    - `goal_reached`: agent has reached it's goal in this episode. It can't perform any more actions until the interaction is resetted.
-    - `blocked`: agent has been blocked. It can't perform any more actions until the interaction is resetted.
-    - `max_steps`: agent has reached it's maximum allowed steps. It can't perform any more actions until the interaction is resetted.
-
-
+`self._agent_last_action`: last Action per agent
+`self._agent_statuses`: status of each agent. One of AgentStatus
 `self._agent_rewards`: dictionary of final reward of each agent in the current episod. Only agent's which can't participate in the ongoing episode are listed.
-`self._agent_trajectories`: complete trajectories for each agent in the ongoing episode
-
-## The format of the messages to the agents is
-    {
-    "to_agent": address of client, 
-    "status": {
-        "#players": number of players,
-        "running": true or false,
-        "time": time in game,
-        } ,
-    "message": Generic text messages (optional),
-    "state": (optional) {
-        "observation": observation_object,
-        "ended": if the game ended or not,
-        "reason": reason for ending
-    }
-    }
+`self._agent_trajectories`: complete trajectories for each agent in the ongoing episode