Merge pull request #500 from LLNL/abmarl-473-predator-prey-resources

Abmarl 473 predator prey resources

abmarl/examples/__init__.py

@@ -5,3 +5,4 @@
 from .sim import EmptyABS, MultiAgentSim, MultiAgentGymSpacesSim, \
     MultiAgentContinuousGymSpaceSim, MultiAgentSameSpacesSim
 from .sim import ReachTheTargetSim, RunningAgent, TargetAgent, BarrierAgent
+from .sim import PredatorPreyResourcesSim, ResourceAgent, PreyAgent, PredatorAgent

abmarl/examples/sim/__init__.py

@@ -6,3 +6,6 @@
     MultiAgentContinuousGymSpaceSim, MultiAgentSameSpacesSim
 from .multi_agent_grid_sim import MultiAgentGridSim
 from .reach_the_target import ReachTheTargetSim, RunningAgent, TargetAgent, BarrierAgent
+from .predator_prey_resources import (
+    PredatorPreyResourcesSim, ResourceAgent, PreyAgent, PredatorAgent
+)

abmarl/examples/sim/predator_prey_resources.py

@@ -0,0 +1,111 @@
+
+from abmarl.sim import Agent
+from abmarl.sim.gridworld.agent import MovingAgent, AttackingAgent, \
+    GridObservingAgent, HealthAgent
+from abmarl.sim.gridworld.smart import SmartGridWorldSimulation
+from abmarl.sim.gridworld.actor import MoveActor, BinaryAttackActor
+
+
+class ResourceAgent(HealthAgent):
+    def __init__(
+        self,
+        encoding=1,
+        render_shape='s',
+        render_color='g',
+        **kwargs
+    ):
+        super().__init__(
+            encoding=encoding,
+            render_shape=render_shape,
+            render_color=render_color,
+            **kwargs
+        )
+
+
+class PreyAgent(HealthAgent, MovingAgent, AttackingAgent, GridObservingAgent):
+    def __init__(
+        self,
+        encoding=2,
+        render_color='b',
+        move_range=1,
+        attack_range=1,
+        attack_strength=1,
+        attack_accuracy=1,
+        view_range=3,
+        **kwargs
+    ):
+        super().__init__(
+            encoding=encoding,
+            render_color=render_color,
+            move_range=move_range,
+            attack_range=attack_range,
+            attack_strength=attack_strength,
+            attack_accuracy=attack_accuracy,
+            view_range=view_range,
+            **kwargs
+        )
+
+
+class PredatorAgent(HealthAgent, MovingAgent, AttackingAgent, GridObservingAgent):
+    def __init__(
+        self,
+        encoding=3,
+        render_color='r',
+        render_shape='d',
+        move_range=1,
+        attack_range=2,
+        attack_strength=1,
+        attack_accuracy=1,
+        view_range=3,
+        **kwargs
+    ):
+        super().__init__(
+            encoding=encoding,
+            render_color=render_color,
+            render_shape=render_shape,
+            move_range=move_range,
+            attack_range=attack_range,
+            attack_strength=attack_strength,
+            attack_accuracy=attack_accuracy,
+            view_range=view_range,
+            **kwargs
+        )
+
+
+class PredatorPreyResourcesSim(SmartGridWorldSimulation):
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+
+        self.move_actor = MoveActor(**kwargs)
+        self.attack_actor = BinaryAttackActor(**kwargs)
+
+        self.finalize()
+
+    def step(self, action_dict, **kwargs):
+        # Process the attacks
+        for agent_id, action in action_dict.items():
+            agent = self.agents[agent_id]
+            if agent.active:
+                attack_status, attacked_agents = \
+                    self.attack_actor.process_action(agent, action, **kwargs)
+                if attack_status: # Attack was attempted
+                    if not attacked_agents: # Attack failed
+                        self.rewards[agent_id] -= 0.1
+                    else:
+                        for attacked_agent in attacked_agents:
+                            if not attacked_agent.active: # Agent has died
+                                self.rewards[agent_id] += 1
+                                if isinstance(attacked_agent, Agent):
+                                    self.rewards[attacked_agent.id] -= 1
+
+        # Process the moves
+        for agent_id, action in action_dict.items():
+            agent = self.agents[agent_id]
+            if agent.active:
+                move_result = self.move_actor.process_action(agent, action, **kwargs)
+                if not move_result:
+                    self.rewards[agent_id] -= 0.1
+
+        # Entropy penalty
+        for agent_id in action_dict:
+            self.rewards[agent_id] -= 0.01

examples/rllib_predator_prey_resources.py

@@ -0,0 +1,96 @@
+
+from abmarl.examples import ResourceAgent, PreyAgent, PredatorAgent, PredatorPreyResourcesSim
+from abmarl.managers import AllStepManager
+from abmarl.external import MultiAgentWrapper
+
+resources = {
+    f'resource_{i}': ResourceAgent(id=f'resource_{i}') for i in range(11)
+}
+prey = {
+    f'prey_{i}': PreyAgent(id=f'prey_{i}') for i in range(5)
+}
+predators = {
+    f'predator_{i}': PredatorAgent(id=f'predator_{i}') for i in range(2)
+}
+agents = {**resources, **prey, **predators}
+
+overlap_map = {
+    1: {2, 3},
+    2: {1, 2, 3},
+    3: {1, 2},
+}
+attack_map = {
+    2: {1},
+    3: {2}
+}
+sim = MultiAgentWrapper(
+    AllStepManager(
+        PredatorPreyResourcesSim.build_sim(
+            20, 20,
+            agents=agents,
+            overlapping=overlap_map,
+            attack_mapping=attack_map,
+            target_mapping=attack_map,
+            states={'PositionState', 'HealthState'},
+            observers={'PositionCenteredEncodingObserver'},
+            dones={'ActiveDone', 'TargetEncodingInactiveDone'}
+        ),
+        randomize_action_input=True
+    )
+)
+
+
+sim_name = "PredatorPreyResources"
+from ray.tune.registry import register_env
+register_env(sim_name, lambda sim_config: sim)
+
+
+policies = {
+    'prey': (None, prey['prey_0'].observation_space, prey['prey_0'].action_space, {}),
+    'predator': (
+        None, predators['predator_0'].observation_space, predators['predator_0'].action_space, {}
+    ),
+}
+
+
+def policy_mapping_fn(agent_id):
+    if agents[agent_id].encoding == 1:
+        return 'prey'
+    if agents[agent_id].encoding == 2:
+        return 'predator'
+
+
+# Experiment parameters
+params = {
+    'experiment': {
+        'title': f'{sim_name}',
+        'sim_creator': lambda config=None: sim,
+    },
+    'ray_tune': {
+        'run_or_experiment': 'PPO',
+        'checkpoint_freq': 50,
+        'checkpoint_at_end': True,
+        'stop': {
+            'episodes_total': 20_000,
+        },
+        'verbose': 2,
+        'local_dir': 'output_dir',
+        'config': {
+            # --- Simulation ---
+            'disable_env_checking': False,
+            'env': sim_name,
+            'horizon': 200,
+            'env_config': {},
+            # --- Multiagent ---
+            'multiagent': {
+                'policies': policies,
+                'policy_mapping_fn': policy_mapping_fn,
+            },
+            # --- Parallelism ---
+            # Number of workers per experiment: int
+            "num_workers": 7,
+            # Number of simulations that each worker starts: int
+            "num_envs_per_worker": 1, # This must be 1 because we are not "threadsafe"
+        },
+    }
+}