[Feature] More VMAS tasks (#88)

* amend

* empty

* amend

* amend

* amend

* amend

* amend

* amend

* change examples

* amend

* amend

* amend

* add

benchmarl/conf/task/vmas/ball_passage.yaml

@@ -0,0 +1,8 @@
+defaults:
+  - _self_
+  - vmas_ball_passage_config
+
+max_steps: 500
+n_passages: 1
+fixed_passage: False
+random_start_angle: True

benchmarl/conf/task/vmas/ball_trajectory.yaml

@@ -0,0 +1,6 @@
+defaults:
+  - _self_
+  - vmas_ball_trajectory_config
+
+max_steps: 100
+joints: True

benchmarl/conf/task/vmas/buzz_wire.yaml

@@ -0,0 +1,7 @@
+defaults:
+  - _self_
+  - vmas_buzz_wire_config
+
+max_steps: 100
+random_start_angle: True
+collision_reward: -10

benchmarl/conf/task/vmas/discovery.yaml

@@ -0,0 +1,12 @@
+defaults:
+  - _self_
+  - vmas_discovery_config
+
+max_steps: 100
+n_agents: 5
+n_targets: 7
+lidar_range: 0.35
+covering_range: 0.25
+agents_per_target: 2
+targets_respawn: True
+shared_reward: True

benchmarl/conf/task/vmas/flocking.yaml

@@ -0,0 +1,8 @@
+defaults:
+  - _self_
+  - vmas_flocking_config
+
+max_steps: 100
+n_agents: 4
+n_targets: 5
+collision_reward: -0.1

benchmarl/conf/task/vmas/joint_passage.yaml

@@ -0,0 +1,14 @@
+defaults:
+  - _self_
+  - vmas_joint_passage_config
+
+max_steps: 500
+n_passages: 1
+fixed_passage: True
+joint_length: 0.5
+random_start_angle: True
+random_goal_angle: True
+observe_joint_angle: False
+asym_package: True
+mass_ratio: 5
+mass_position: 0.75

benchmarl/conf/task/vmas/joint_passage_size.yaml

@@ -0,0 +1,10 @@
+defaults:
+  - _self_
+  - vmas_joint_passage_size_config
+
+max_steps: 500
+n_passages: 3
+fixed_passage: False
+random_start_angle: False
+random_goal_angle: False
+observe_joint_angle: False

benchmarl/conf/task/vmas/multi_give_way.yaml

@@ -0,0 +1,6 @@
+defaults:
+  - _self_
+  - vmas_multi_give_way_config
+
+max_steps: 200
+agent_collision_penalty: -0.1

benchmarl/conf/task/vmas/passage.yaml

@@ -0,0 +1,7 @@
+defaults:
+  - _self_
+  - vmas_passage_config
+
+max_steps: 500
+n_passages: 1
+shared_reward: True

benchmarl/environments/__init__.py

@@ -21,7 +21,7 @@
 
 
 from .pettingzoo.multiwalker import TaskConfig as MultiwalkerConfig
-from .pettingzoo.simple_adverasary import TaskConfig as SimpleAdversaryConfig
+from .pettingzoo.simple_adversary import TaskConfig as SimpleAdversaryConfig
 from .pettingzoo.simple_crypto import TaskConfig as SimpleCryptoConfig
 from .pettingzoo.simple_push import TaskConfig as SimplePushConfig
 from .pettingzoo.simple_reference import TaskConfig as SimpleReferenceConfig
@@ -34,13 +34,22 @@
 from .pettingzoo.waterworld import TaskConfig as WaterworldConfig
 
 from .vmas.balance import TaskConfig as BalanceConfig
+from .vmas.ball_passage import TaskConfig as BallPassageConfig
+from .vmas.ball_trajectory import TaskConfig as BallTrajectoryConfig
+from .vmas.buzz_wire import TaskConfig as BuzzWireConfig
+from .vmas.discovery import TaskConfig as DiscoveryConfig
 from .vmas.dispersion import TaskConfig as DispersionConfig
 from .vmas.dropout import TaskConfig as DropoutConfig
+from .vmas.flocking import TaskConfig as FlockingConfig
 from .vmas.give_way import TaskConfig as GiveWayConfig
+from .vmas.joint_passage import TaskConfig as JointPassageConfig
+from .vmas.joint_passage_size import TaskConfig as JointPassageSizeConfig
+from .vmas.multi_give_way import TaskConfig as MultiGiveWayConfig
 from .vmas.navigation import TaskConfig as NavigationConfig
+from .vmas.passage import TaskConfig as PassageConfig
 from .vmas.reverse_transport import TaskConfig as ReverseTransportConfig
 from .vmas.sampling import TaskConfig as SamplingConfig
-from .vmas.simple_adverasary import TaskConfig as VmasSimpleAdversaryConfig
+from .vmas.simple_adversary import TaskConfig as VmasSimpleAdversaryConfig
 from .vmas.simple_crypto import TaskConfig as VmasSimpleCryptoConfig
 from .vmas.simple_push import TaskConfig as VmasSimplePushConfig
 from .vmas.simple_reference import TaskConfig as VmasSimpleReferenceConfig
@@ -52,7 +61,6 @@
 from .vmas.wheel import TaskConfig as WheelConfig
 from .vmas.wind_flocking import TaskConfig as WindFlockingConfig
 
-
 # This is a registry mapping task config schemas names to their python dataclass
 # It is used by hydra to validate loaded configs.
 # You will see the "envname_taskname_config" strings in the hydra defaults at the top of yaml files.
@@ -66,8 +74,17 @@
     "vmas_wheel_config": WheelConfig,
     "vmas_dispersion_config": DispersionConfig,
     "vmas_give_way_config": GiveWayConfig,
+    "vmas_multi_give_way_config": MultiGiveWayConfig,
+    "vmas_passage_config": PassageConfig,
+    "vmas_joint_passage_config": JointPassageConfig,
+    "vmas_joint_passage_size_config": JointPassageSizeConfig,
+    "vmas_ball_passage_config": BallPassageConfig,
+    "vmas_buzz_wire_config": BuzzWireConfig,
+    "vmas_ball_trajectory_config": BallTrajectoryConfig,
+    "vmas_flocking_config": FlockingConfig,
     "vmas_wind_flocking_config": WindFlockingConfig,
     "vmas_dropout_config": DropoutConfig,
+    "vmas_discovery_config": DiscoveryConfig,
     "vmas_simple_adversary_config": VmasSimpleAdversaryConfig,
     "vmas_simple_crypto_config": VmasSimpleCryptoConfig,
     "vmas_simple_push_config": VmasSimplePushConfig,

benchmarl/environments/pettingzoo/simple_reference.py

@@ -11,3 +11,4 @@
 class TaskConfig:
     task: str = MISSING
     max_cycles: int = MISSING
+    local_ratio: float = MISSING

benchmarl/environments/vmas/ball_passage.py

@@ -0,0 +1,15 @@
+#  Copyright (c) Meta Platforms, Inc. and affiliates.
+#
+#  This source code is licensed under the license found in the
+#  LICENSE file in the root directory of this source tree.
+#
+
+from dataclasses import dataclass, MISSING
+
+
+@dataclass
+class TaskConfig:
+    max_steps: int = MISSING
+    n_passages: int = MISSING
+    fixed_passage: bool = MISSING
+    random_start_angle: bool = MISSING

benchmarl/environments/vmas/ball_trajectory.py

@@ -0,0 +1,13 @@
+#  Copyright (c) Meta Platforms, Inc. and affiliates.
+#
+#  This source code is licensed under the license found in the
+#  LICENSE file in the root directory of this source tree.
+#
+
+from dataclasses import dataclass, MISSING
+
+
+@dataclass
+class TaskConfig:
+    max_steps: int = MISSING
+    joints: bool = MISSING

benchmarl/environments/vmas/buzz_wire.py

@@ -0,0 +1,14 @@
+#  Copyright (c) Meta Platforms, Inc. and affiliates.
+#
+#  This source code is licensed under the license found in the
+#  LICENSE file in the root directory of this source tree.
+#
+
+from dataclasses import dataclass, MISSING
+
+
+@dataclass
+class TaskConfig:
+    max_steps: int = MISSING
+    random_start_angle: bool = MISSING
+    collision_reward: float = MISSING

benchmarl/environments/vmas/common.py

@@ -24,9 +24,18 @@ class VmasTask(Task):
     REVERSE_TRANSPORT = None
     WHEEL = None
     DISPERSION = None
+    MULTI_GIVE_WAY = None
     DROPOUT = None
     GIVE_WAY = None
     WIND_FLOCKING = None
+    PASSAGE = None
+    JOINT_PASSAGE = None
+    JOINT_PASSAGE_SIZE = None
+    BALL_PASSAGE = None
+    BALL_TRAJECTORY = None
+    BUZZ_WIRE = None
+    FLOCKING = None
+    DISCOVERY = None
     SIMPLE_ADVERSARY = None
     SIMPLE_CRYPTO = None
     SIMPLE_PUSH = None

benchmarl/environments/vmas/discovery.py

@@ -0,0 +1,19 @@
+#  Copyright (c) Meta Platforms, Inc. and affiliates.
+#
+#  This source code is licensed under the license found in the
+#  LICENSE file in the root directory of this source tree.
+#
+
+from dataclasses import dataclass, MISSING
+
+
+@dataclass
+class TaskConfig:
+    max_steps: int = MISSING
+    n_agents: int = MISSING
+    n_targets: int = MISSING
+    lidar_range: float = MISSING
+    covering_range: float = MISSING
+    agents_per_target: int = MISSING
+    targets_respawn: bool = MISSING
+    shared_reward: bool = MISSING

benchmarl/environments/vmas/flocking.py

@@ -0,0 +1,15 @@
+#  Copyright (c) Meta Platforms, Inc. and affiliates.
+#
+#  This source code is licensed under the license found in the
+#  LICENSE file in the root directory of this source tree.
+#
+
+from dataclasses import dataclass, MISSING
+
+
+@dataclass
+class TaskConfig:
+    max_steps: int = MISSING
+    n_agents: int = MISSING
+    n_targets: int = MISSING
+    collision_reward: float = MISSING

benchmarl/environments/vmas/joint_passage.py

@@ -0,0 +1,21 @@
+#  Copyright (c) Meta Platforms, Inc. and affiliates.
+#
+#  This source code is licensed under the license found in the
+#  LICENSE file in the root directory of this source tree.
+#
+
+from dataclasses import dataclass, MISSING
+
+
+@dataclass
+class TaskConfig:
+    max_steps: int = MISSING
+    n_passages: int = MISSING
+    fixed_passage: bool = MISSING
+    joint_length: float = MISSING
+    random_start_angle: bool = MISSING
+    random_goal_angle: bool = MISSING
+    observe_joint_angle: bool = MISSING
+    asym_package: bool = MISSING
+    mass_ratio: float = MISSING
+    mass_position: float = MISSING

benchmarl/environments/vmas/joint_passage_size.py

@@ -0,0 +1,17 @@
+#  Copyright (c) Meta Platforms, Inc. and affiliates.
+#
+#  This source code is licensed under the license found in the
+#  LICENSE file in the root directory of this source tree.
+#
+
+from dataclasses import dataclass, MISSING
+
+
+@dataclass
+class TaskConfig:
+    max_steps: int = MISSING
+    n_passages: int = MISSING
+    fixed_passage: bool = MISSING
+    random_start_angle: bool = MISSING
+    random_goal_angle: bool = MISSING
+    observe_joint_angle: bool = MISSING

benchmarl/environments/vmas/multi_give_way.py

@@ -0,0 +1,13 @@
+#  Copyright (c) Meta Platforms, Inc. and affiliates.
+#
+#  This source code is licensed under the license found in the
+#  LICENSE file in the root directory of this source tree.
+#
+
+from dataclasses import dataclass, MISSING
+
+
+@dataclass
+class TaskConfig:
+    max_steps: int = MISSING
+    agent_collision_penalty: float = MISSING

benchmarl/environments/vmas/passage.py

@@ -0,0 +1,14 @@
+#  Copyright (c) Meta Platforms, Inc. and affiliates.
+#
+#  This source code is licensed under the license found in the
+#  LICENSE file in the root directory of this source tree.
+#
+
+from dataclasses import dataclass, MISSING
+
+
+@dataclass
+class TaskConfig:
+    max_steps: int = MISSING
+    n_passages: int = MISSING
+    shared_reward: bool = MISSING