steffen/marl-factory-grid
1
0
Fork 0
mirror of https://github.com/illiumst/marl-factory-grid.git synced 2025-09-18 00:21:58 +02:00
Code Issues Packages Projects Releases Wiki Activity

major redesign ob observations and entittes

Browse Source
This commit is contained in:
Steffen Illium
2023-06-09 14:04:17 +02:00
parent 901fbcbc32
commit c552c35f66
161 changed files with 4458 additions and 4163 deletions
Download Patch File Download Diff File Expand all files Collapse all files

111
quickstart/all_test_config.yaml Normal file
View File

@@ -0,0 +1,111 @@
---
General:
level_name: large
env_seed: 69
verbose: !!bool False
pomdp_r: 3
individual_rewards: !!bool True
Entities:
Defaults: {}
DirtPiles:
initial_dirt_ratio: 0.01 # On INIT, on max how many tiles does the dirt spawn in percent.
dirt_spawn_r_var: 0.5 # How much does the dirt spawn amount vary?
initial_amount: 1
max_local_amount: 3 # Max dirt amount per tile.
max_global_amount: 30 # Max dirt amount in the whole environment.
Doors:
closed_on_init: True
auto_close_interval: 10
indicate_area: False
Batteries: {}
ChargePods: {}
Destinations: {}
ReachedDestinations: {}
Items: {}
Inventories: {}
DropOffLocations: {}
Agents:
Wolfgang:
Actions:
- Move8
- DoorUse
- CleanUp
Observations:
- Self
- Placeholder
- Walls
- DirtPiles
- Placeholder
- Doors
- Doors
Bjoern:
Actions:
# Move4, Noop
- Move8
- DoorUse
- ItemAction
Observations:
- Defaults
- Combined:
- Other
- Walls
- Items
- Inventory
Karl-Heinz:
Actions:
- Move8
- DoorUse
Observations:
# Wall, Only Other Agents
- Defaults
- Combined:
- Other
- Self
- Walls
- Doors
- Destinations
Manfred:
Actions:
- Move8
- ItemAction
- DoorUse
- CleanUp
- DestAction
- BtryCharge
Observations:
- Defaults
- Battery
- Destinations
- DirtPiles
- Doors
- Items
- Inventory
- DropOffLocations
Rules:
Defaults: {}
Collision:
done_at_collisions: !!bool False
DirtRespawnRule:
spawn_freq: 15
DirtSmearOnMove:
smear_amount: 0.12
DoorAutoClose: {}
DirtAllCleanDone: {}
Btry: {}
BtryDoneAtDischarge: {}
DestinationReach: {}
DestinationSpawn: {}
DestinationDone: {}
ItemRules: {}
Assets:
- Defaults
- Dirt
- Door
- Machine
- Item
- Destination
- DropOffLocation
- Chargepod

2
quickstart/combine_and_monitor_rerun.py
View File

@@ -134,7 +134,7 @@ if __name__ == '__main__':
else:
assert combined_env_kwargs[key] == val, "Check the combinations you try to make!"
# Update and combine all kwargs to account for multiple agents etc.
# Update and combine all kwargs to account for multiple agent etc.
# We cannot capture all configuration cases!
for key, val in factory_kwargs.items():
if key not in combined_env_kwargs:

68
quickstart/default_config.yaml Normal file
View File

@@ -0,0 +1,68 @@
---
General:
level_name: rooms
env_seed: 69
verbose: !!bool False
pomdp_r: 5
individual_rewards: !!bool True
Entities:
Defaults: {}
DirtPiles:
initial_dirt_ratio: 0.3 # On INIT, on max how many tiles does the dirt spawn in percent.
dirt_spawn_r_var: 0.05 # How much does the dirt spawn amount vary?
initial_amount: 3
max_local_amount: 5 # Max dirt amount per tile.
max_global_amount: 20 # Max dirt amount in the whole environment.
Doors:
closed_on_init: True
auto_close_interval: 10
indicate_area: False
Agents:
Wolfgang:
Actions:
- Move8
- Noop
- DoorUse
- CleanUp
Observations:
- Self
- Placeholder
- Walls
- DirtPiles
- Placeholder
- Doors
- Doors
Björn:
Actions:
# Move4, Noop
- Move4
- DoorUse
- CleanUp
Observations:
- Defaults
- Combined
Jürgen:
Actions:
# Move4, Noop
- Defaults
- DoorUse
- CleanUp
Observations:
- Walls
- Placeholder
- Agent[Björn]
Rules:
Defaults: {}
Collision:
done_at_collisions: !!bool False
DirtRespawnRule:
spawn_freq: 5
DirtSmearOnMove:
smear_amount: 0.12
DoorAutoClose: {}
DirtAllCleanDone: {}
Assets:
- Defaults
- Dirt
- Doors

8
quickstart/single_agent_train_battery_target_env.py
View File

@@ -55,7 +55,7 @@ if __name__ == '__main__':
# Define a global studi save path
start_time = int(time.time())
study_root_path = Path(__file__).parent.parent / 'study_out' / f'{Path(__file__).stem}_{start_time}'
# Create an identifier, which is unique for every combination and easy to read in filesystem
# Create an _identifier, which is unique for every combination and easy to read in filesystem
identifier = f'{model_class.__name__}_{env_class.__name__}_{start_time}'
exp_path = study_root_path / identifier
@@ -67,9 +67,9 @@ if __name__ == '__main__':
# 'ObservationProperties' are for specifying how the agent sees the env.
obs_props = ObservationProperties(render_agents=AgentRenderOptions.NOT, # Agents won`t be shown in the obs at all
omit_agent_self=True, # This is default
additional_agent_placeholder=None, # We will not take care of future agents
additional_agent_placeholder=None, # We will not take care of future agent
frames_to_stack=3, # To give the agent a notion of time
pomdp_r=2 # the agents view-radius
pomdp_r=2 # the agent view-radius
)
# 'MovementProperties' are for specifying how the agent is allowed to move in the env.
move_props = MovementProperties(allow_diagonal_movement=True, # Euclidean style (vertices)
@@ -169,7 +169,7 @@ if __name__ == '__main__':
# Evaluation starts here #####################################################
# First Iterate over every model and monitor "as trained"
print('Start Measurement Tracking')
# For trained policy in study_root_path / identifier
# For trained policy in study_root_path / _identifier
for policy_path in [x for x in exp_path.iterdir() if x.is_dir()]:
# retrieve model class

8
quickstart/single_agent_train_dest_env.py
View File

@@ -54,7 +54,7 @@ if __name__ == '__main__':
# Define a global studi save path
start_time = int(time.time())
study_root_path = Path(__file__).parent.parent / 'study_out' / f'{Path(__file__).stem}_{start_time}'
# Create an identifier, which is unique for every combination and easy to read in filesystem
# Create an _identifier, which is unique for every combination and easy to read in filesystem
identifier = f'{model_class.__name__}_{env_class.__name__}_{start_time}'
exp_path = study_root_path / identifier
@@ -66,9 +66,9 @@ if __name__ == '__main__':
# 'ObservationProperties' are for specifying how the agent sees the env.
obs_props = ObservationProperties(render_agents=AgentRenderOptions.NOT, # Agents won`t be shown in the obs at all
omit_agent_self=True, # This is default
additional_agent_placeholder=None, # We will not take care of future agents
additional_agent_placeholder=None, # We will not take care of future agent
frames_to_stack=3, # To give the agent a notion of time
pomdp_r=2 # the agents view-radius
pomdp_r=2 # the agent view-radius
)
# 'MovementProperties' are for specifying how the agent is allowed to move in the env.
move_props = MovementProperties(allow_diagonal_movement=True, # Euclidean style (vertices)
@@ -159,7 +159,7 @@ if __name__ == '__main__':
# Evaluation starts here #####################################################
# First Iterate over every model and monitor "as trained"
print('Start Measurement Tracking')
# For trained policy in study_root_path / identifier
# For trained policy in study_root_path / _identifier
for policy_path in [x for x in exp_path.iterdir() if x.is_dir()]:
# retrieve model class

8
quickstart/single_agent_train_dirt_env.py
View File

@@ -54,7 +54,7 @@ if __name__ == '__main__':
# Define a global studi save path
start_time = int(time.time())
study_root_path = Path(__file__).parent.parent / 'study_out' / f'{Path(__file__).stem}_{start_time}'
# Create an identifier, which is unique for every combination and easy to read in filesystem
# Create an _identifier, which is unique for every combination and easy to read in filesystem
identifier = f'{model_class.__name__}_{env_class.__name__}_{start_time}'
exp_path = study_root_path / identifier
@@ -66,9 +66,9 @@ if __name__ == '__main__':
# 'ObservationProperties' are for specifying how the agent sees the env.
obs_props = ObservationProperties(render_agents=AgentRenderOptions.NOT, # Agents won`t be shown in the obs at all
omit_agent_self=True, # This is default
additional_agent_placeholder=None, # We will not take care of future agents
additional_agent_placeholder=None, # We will not take care of future agent
frames_to_stack=3, # To give the agent a notion of time
pomdp_r=2 # the agents' view-radius
pomdp_r=2 # the agent' view-radius
)
# 'MovementProperties' are for specifying how the agent is allowed to move in the env.
move_props = MovementProperties(allow_diagonal_movement=True, # Euclidean style (vertices)
@@ -161,7 +161,7 @@ if __name__ == '__main__':
# Evaluation starts here #####################################################
# First Iterate over every model and monitor "as trained"
print('Start Measurement Tracking')
# For trained policy in study_root_path / identifier
# For trained policy in study_root_path / _identifier
for policy_path in [x for x in exp_path.iterdir() if x.is_dir()]:
# retrieve model class

8
quickstart/single_agent_train_item_env.py
View File

@@ -54,7 +54,7 @@ if __name__ == '__main__':
# Define a global studi save path
start_time = int(time.time())
study_root_path = Path(__file__).parent.parent / 'study_out' / f'{Path(__file__).stem}_{start_time}'
# Create an identifier, which is unique for every combination and easy to read in filesystem
# Create an _identifier, which is unique for every combination and easy to read in filesystem
identifier = f'{model_class.__name__}_{env_class.__name__}_{start_time}'
exp_path = study_root_path / identifier
@@ -65,9 +65,9 @@ if __name__ == '__main__':
# 'ObservationProperties' are for specifying how the agent sees the env.
obs_props = ObservationProperties(render_agents=AgentRenderOptions.NOT, # Agents won`t be shown in the obs at all
omit_agent_self=True, # This is default
additional_agent_placeholder=None, # We will not take care of future agents
additional_agent_placeholder=None, # We will not take care of future agent
frames_to_stack=3, # To give the agent a notion of time
pomdp_r=2 # the agents view-radius
pomdp_r=2 # the agent view-radius
)
# 'MovementProperties' are for specifying how the agent is allowed to move in the env.
move_props = MovementProperties(allow_diagonal_movement=True, # Euclidean style (vertices)
@@ -157,7 +157,7 @@ if __name__ == '__main__':
# Evaluation starts here #####################################################
# First Iterate over every model and monitor "as trained"
print('Start Measurement Tracking')
# For trained policy in study_root_path / identifier
# For trained policy in study_root_path / _identifier
for policy_path in [x for x in exp_path.iterdir() if x.is_dir()]:
# retrieve model class