steffen/marl-factory-grid
1
0
Fork 0
mirror of https://github.com/illiumst/marl-factory-grid.git synced 2025-07-08 02:21:36 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'origin/marl_refactor' into marl_refactor

Browse Source
This commit is contained in:
Chanumask
2024-05-21 11:40:22 +02:00
parent 13ea9d25c9 4571dc1cd1
commit c7f2bbfbac
9 changed files with 64 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
marl_factory_grid/algorithms/marl/a2c_dirt.py
View File

@ -384,6 +384,14 @@ class A2C:
obs[0][1][x][y] = 1
print("Missing agent position")
def get_all_cleaned_dirt_piles(self, dirt_piles_positions, cleaned_dirt_piles):
meta_cleaned_dirt_piles = {pos: False for pos in dirt_piles_positions}
for agent_idx in range(self.n_agents):
for (pos, cleaned) in cleaned_dirt_piles[agent_idx].items():
if cleaned:
meta_cleaned_dirt_piles[pos] = True
return meta_cleaned_dirt_piles
def handle_dirt(self, env, cleaned_dirt_piles, ordered_dirt_piles, target_pile, indices, reward, done):
# Check if agent moved on field with dirt. If that is the case collect dirt automatically
agent_positions = [env.state.moving_entites[agent_idx].pos for agent_idx in range(self.n_agents)]
@ -428,12 +436,7 @@ class A2C:
done = True
elif self.cfg[nms.ALGORITHM]["pile_all_done"] == "shared":
# End episode if both agents together have cleaned all dirt piles
meta_cleaned_dirt_piles = {pos: False for pos in dirt_piles_positions}
for agent_idx in range(self.n_agents):
for (pos, cleaned) in cleaned_dirt_piles[agent_idx].items():
if cleaned:
meta_cleaned_dirt_piles[pos] = True
if all(meta_cleaned_dirt_piles.values()):
if all(self.get_all_cleaned_dirt_piles(dirt_piles_positions, cleaned_dirt_piles).values()):
done = True
return reward, done

2
marl_factory_grid/algorithms/marl/configs/dirt_quadrant_config.yaml
View File

@ -24,7 +24,7 @@ algorithm:
entropy_coef: 0.01
vf_coef: 0.05
n_steps: 0 # How much experience should be sampled at most (n-TD) until the next value and policy update is performed. Default 0: MC
max_steps: 270000
max_steps: 240000
advantage: "Advantage-AC" # Options: "Advantage-AC", "TD-Advantage-AC", "Reinforce"
pile-order: "fixed" # Options: "fixed", "random", "none", "agents", "dynamic", "smart" (Use "fixed", "random" and "none" for single agent training and the other for multi agent inference)
pile-observability: "single" # Options: "single", "all"

2
marl_factory_grid/configs/custom/MultiAgentConfigs/dirt_quadrant_eval_config.yaml
View File

@ -37,7 +37,7 @@ Agents:
Entities:
DirtPiles:
coords_or_quantity: (9,9), (4,5), (1,1) # (4,7), (2,4), (1, 1) # (1, 1), (2,4), (4,7), (7,9), (9,9) # (1, 1), (1,2), (1,3), (2,4), (2,5), (3,6), (4,7), (5,8), (6,8), (7,9), (8,9), (9,9)
coords_or_quantity: (9,9), (7,9), (4,7), (2,4), (1, 1) # (4,7), (2,4), (1, 1) # (1, 1), (2,4), (4,7), (7,9), (9,9) # (1, 1), (1,2), (1,3), (2,4), (2,5), (3,6), (4,7), (5,8), (6,8), (7,9), (8,9), (9,9)
initial_amount: 0.5 # <1 to ensure that the robot which first attempts to clean this field, can remove the dirt in one action
clean_amount: 1
dirt_spawn_r_var: 0

2
marl_factory_grid/configs/custom/MultiAgentConfigs/two_rooms_one_door_modified_eval_config_emergent.yaml
View File

@ -59,4 +59,4 @@ Rules:
# Done Conditions
#DoneOnAllDirtCleaned:
DoneAtMaxStepsReached:
max_steps: 50
max_steps: 30

22
marl_factory_grid/configs/custom/dirt_quadrant_eval_config.yaml
View File

@ -25,8 +25,15 @@ Agents:
#- Self
#Positions:
#- (9,1)
#- (1,1)
#- (2,4)
#- (4,7)
#- (7,9)
#- (2,4)
#- (4,7)
#- (7,9)
#- (9,9)
#- (4,5)
#- (9,1)
Wolfgang:
Actions:
- Move4
@ -35,12 +42,19 @@ Agents:
- Self
Positions:
- (9,5)
- (9,9)
- (4,5)
#- (1,1)
#- (2,4)
#- (4,7)
#- (7,9)
#- (2,4)
#- (4,7)
#- (7,9)
#- (9,9)
#- (9,5)
Entities:
DirtPiles:
coords_or_quantity: (9,9), (4,5), (1,1) # (4,7), (2,4), (1, 1) # (1, 1), (2,4), (4,7), (7,9), (9,9) # (1, 1), (1,2), (1,3), (2,4), (2,5), (3,6), (4,7), (5,8), (6,8), (7,9), (8,9), (9,9)
coords_or_quantity: (1, 1), (2,4), (4,7), (7,9), (9,9) #(9,9), (7,9), (4,7), (2,4), (1, 1) #(1, 1), (2,4), (4,7), (7,9), (9,9) # (4,7), (2,4), (1, 1) # (1, 1), (2,4), (4,7), (7,9), (9,9) # (1, 1), (1,2), (1,3), (2,4), (2,5), (3,6), (4,7), (5,8), (6,8), (7,9), (8,9), (9,9)
initial_amount: 0.5 # <1 to ensure that the robot which first attempts to clean this field, can remove the dirt in one action
clean_amount: 1
dirt_spawn_r_var: 0

26
marl_factory_grid/configs/custom/dirt_quadrant_train_config.yaml
View File

@ -24,11 +24,17 @@ Agents:
#- Self
#Positions:
#- (9,1)
#- (4,5)
#- (1,1)
#- (4,5)
#- (9,1)
#- (2,4)
#- (4,7)
#- (6,8)
#- (7,9)
#- (2,4)
#- (4,7)
#- (6,8)
#- (7,9)
#- (9,9)
#- (9,1)
Wolfgang:
Actions:
- Move4
@ -37,16 +43,22 @@ Agents:
- Self
Positions:
- (9,5)
- (4,5)
- (1,1)
- (4,5)
- (9,5)
- (2,4)
- (4,7)
- (6,8)
- (7,9)
- (2,4)
- (4,7)
- (6,8)
- (7,9)
- (9,9)
- (9,5)
Entities:
DirtPiles:
coords_or_quantity: (9,9), (1,1), (4,5) # (4,7), (2,4), (1, 1) #(1, 1), (2,4), (4,7), (7,9), (9,9) # (1, 1), (1,2), (1,3), (2,4), (2,5), (3,6), (4,7), (5,8), (6,8), (7,9), (8,9), (9,9)
coords_or_quantity: (1, 1), (2,4), (4,7), (6,8), (7,9), (9,9) # (4,7), (2,4), (1, 1) #(1, 1), (2,4), (4,7), (7,9), (9,9) # (1, 1), (1,2), (1,3), (2,4), (2,5), (3,6), (4,7), (5,8), (6,8), (7,9), (8,9), (9,9)
initial_amount: 0.5 # <1 to ensure that the robot which first attempts to clean this field, can remove the dirt in one action
clean_amount: 1
dirt_spawn_r_var: 0

4
marl_factory_grid/configs/dirt_quadrant.yaml
View File

@ -27,7 +27,7 @@ Agents:
- DirtPiles
- Self
Positions:
- (9,2)
- (9,1)
Reiner:
Actions:
- Move4
@ -43,7 +43,7 @@ Agents:
Entities:
DirtPiles:
coords_or_quantity: (1, 1), (4,5), (9,9)
coords_or_quantity: (1, 1), (2,4), (4,7), (7,9), (9,9)
initial_amount: 0.5 # <1 to ensure that the robot which first attempts to clean this field, can remove the dirt in one action
clean_amount: 1
dirt_spawn_r_var: 0

12
marl_factory_grid/levels/quadrant.txt
View File

@ -1,10 +1,10 @@
###########
#---#######
#-----#####
#------####
#-------###
#--------##
#--------##
#---------#
#---------#
#---------#
#---------#
#---------#
#---------#
#---------#
#---------#
#---------#

14
studies/marl_adapted.py
View File

@ -30,7 +30,7 @@ def single_agent_eval(config_name, run):
agent = A2C(train_cfg, eval_cfg)
print("Evaluation phase")
agent.load_agents(run)
agent.eval_loop(10)
agent.eval_loop(1)
def multi_agent_eval(config_name, runs, emergent_phenomenon=False):
@ -55,7 +55,7 @@ def multi_agent_eval(config_name, runs, emergent_phenomenon=False):
agent = A2C(train_cfg, eval_cfg)
print("Evaluation phase")
agent.load_agents(runs)
agent.eval_loop(10)
agent.eval_loop(1)
def dirt_quadrant_single_agent_training():
@ -70,7 +70,7 @@ def dirt_quadrant_single_agent_eval(agent_name):
if agent_name == "Sigmund":
run = "run0"
elif agent_name == "Wolfgang":
run = "run4"
run = "run1"
single_agent_eval("dirt_quadrant", [run])
@ -82,15 +82,15 @@ def two_rooms_one_door_modified_single_agent_eval(agent_name):
single_agent_eval("two_rooms_one_door_modified", [run])
def dirt_quadrant_multi_agent_eval(emergent_phenomenon):
multi_agent_eval("dirt_quadrant", ["run0", "run1"], emergent_phenomenon)
def dirt_quadrant_5_multi_agent_eval(emergent_phenomenon):
multi_agent_eval("dirt_quadrant", ["run4", "run5"], emergent_phenomenon)
def dirt_quadrant_5_multi_agent_ctde_eval(emergent_phenomenon): # run7 == run4
multi_agent_eval("dirt_quadrant", ["run4", "run7"], emergent_phenomenon)
def two_rooms_one_door_modified_multi_agent_eval(emergent_phenomenon):
multi_agent_eval("two_rooms_one_door_modified", ["run2", "run3"], emergent_phenomenon)
if __name__ == '__main__':
dirt_quadrant_single_agent_training()