# Gneral env. settings.
General:
  # Just the best seed.
  env_seed: 69
  # Each agent receives an inividual Reward.
  individual_rewards: true
  # level file to load from .\levels\.
  level_name: eight_puzzle
  # Partial Observability. 0 = Full Observation.
  pomdp_r: 0
  # Please do not spam me.
  verbose: false
  # Do not touch, WIP
  tests: false

# RL Surrogates
Agents:
  # This defines the name of the agent. UTF-8
  Wolfgang:
    # Section which defines the availabll Actions per Agent
    Actions:
      # Move4 adds 4 actions [`North`, `East`, `South`, `West`]
      Move4:
        # Reward specification which differ from the default.
        # Agent does a valid move in the environment. He actually moves.
        valid_reward: -0.1
        # Agent wants to move, but fails.
        fail_reward:  0
      # NOOP aka agent does not do a thing.
      Noop:
        # The Agent decides to not do anything. Which is always valid.
        valid_reward: 0
        # Does not do anything, just using the same interface.
        fail_reward: 0
    # What the agent wants to see.
    Observations:
      # The agent...
      # sees other agents, but himself.
      - Other
      # wants to see walls
      - Walls
      # sees his associated Destination (singular). Use the Plural for `see all destinations`.
      - Destination
    # You want to have 7 clones, also possible to name them by giving names as list.
    Clones: 7
    # Agents are blocking their grid position from beeing entered by others.
    is_blocking_pos: true
# Apart from agents, which additional endities do you want to load?
Entities:
  # Observable destinations, which can be reached by stepping on the same position. Has additional parameters...
  Destinations:
    # Let them spawn on closed doors and agent positions
    ignore_blocking: true
    # For 8-Puzzle, we need a special spawn rule...
    spawnrule:
      # ...which spawn a single position just underneath an associated agent.
      SpawnDestinationOnAgent: {}  # There are no parameters, so we state empty kwargs.

# This section defines which operations are performed beside agent action.
# Without this section nothing happens, not even Done-condition checks.
# Also, situation based rewards are specidief this way.
Rules:
  ## Utilities
  # This rule defines the collision mechanic, introduces a related DoneCondition and lets you specify rewards.
  # Can be omited/ignored if you do not want to take care of collisions at all.
  #   This does not mean, that agents can not collide, its just ignored.
  WatchCollisions:
    reward: 0
    done_at_collisions: false

  # In 8 Puzzle, do not randomize the start positions, rather move a random agent onto the single free position n-times.
  DoRandomInitialSteps:
    # How many times?
    random_steps: 2

  ## Done Conditions
  # Maximum steps per episode. There is no reward for failing.
  DoneAtMaxStepsReached:
    # After how many steps should the episode end?
    max_steps: 200

  # For 8 Puzzle we need a done condition that checks whether destinations have been reached, so...
  DoneAtDestinationReach:
    # On every step, should there be a reward for agets that reach their associated destination? No!
    dest_reach_reward: 0  # Do not touch. This is usefull in other settings!
    # Reward should only be given when all destiantions are reached in parallel!
    condition: "simultanious"
    # Reward if this is the case. Granted to each agent when all agents are at their target position simultaniously.
    reward_at_done: 1