separate agents from scenario

Author: zrysnd

multiagent/scenarios/race.py

@@ -1,6 +1,4 @@
 import numpy as np
-import random
-from datetime import datetime
 from multiagent.coreRace import World, Agent, Landmark
 from multiagent.scenario import BaseScenario
 
@@ -11,15 +9,19 @@ class Scenario(BaseScenario):
  def __init__(self):
  super(Scenario, self).__init__()
  self.agentsToLandMarks = {}
- self.numberOfAgents=8
- self.laststate=0
+ self.numberOfAgents=6
+ self.allCooperateForce = 1.0
+ self.allCheatPunishment = -1.0
+ self.cheaterGain = 2.0
+ self.robbedCooperator = 0.0
+
  def make_world(self):
  world = World()
  # add agents
 
  world.agents = [Agent() for i in range(self.numberOfAgents)]
  for i, agent in enumerate(world.agents):
- #print(i)
+ print(i)
  agent.name = 'agent %d' % i
  agent.collide = False
  agent.silent = True
@@ -31,7 +33,7 @@ def make_world(self):
  landmark.movable = False
  #fill in the dictionary
  for i in range(self.numberOfAgents):
- self.agentsToLandMarks.update({ world.agents[i]: world.landmarks[i] })
+  self.agentsToLandMarks.update({ world.agents[i]: world.landmarks[i] })
 
  # make initial conditions
  self.reset_world(world)
@@ -69,61 +71,39 @@ def reward(self, agent, world):
  cheat-cheat: all stay, all cooperate: all move as 1.0, cheat cooperate: cheater moves more.
  '''
 
- random.seed( datetime.now())
+
  def agent_cheated(agent):
  if agent.action.u[1] > 0.0:
  return True
  return False
- 
+
  reward_n = []
  numOfCheaters = 0
  numOfAgents = 0
- # agent part code
+ # if(agent_cheated(world.agents[1])):
+ # print(world.agents[1].action.u)
  for i, thisAgent in enumerate(world.agents):
- j=random.randint(1,10)
- if(i==0):
- if(j>=8):
- thisAgent.action.u[1]=1
- if(i==1):
- if(j>=6):
- thisAgent.action.u[1]=1
- if(i==2):
- if(j>=4):
- thisAgent.action.u[1]=1
- if(i==3):
- thisAgent.action.u[1]=1
- if(i==4):
- thisAgent.action.u[1]=1
- if(i==5):
- thisAgent.action.u[1]=1 
- #for i, thisAgent in enumerate(world.agents):
- #print(thisAgent.action.u[1])
- for i, thisAgent in enumerate(world.agents):
- # if i == 0:
+ # if i == 2:
+ # print(thisAgent.action.u)
  # print(i)
  # print(agent.action.u)
  numOfAgents += 1
  thisAgent.action.u[0] = 0.0 #invalidate horizontal action
  if agent_cheated(thisAgent): #it's trying to move up
  numOfCheaters += 1
- #print(numOfCheaters)
+ # print("agent " + str(i) + "cheated" )
+ # print(numOfCheaters)
 
  for i, thisAgent in enumerate(world.agents):
  if numOfCheaters > numOfAgents//2:
- 
- if thisAgent.action.u[1] == 1:
- thisAgent.action.u[1] = -0.01 # all cheat: move back ward
- if thisAgent.action.u[1] == 0:
- thisAgent.action.u[1] = 0.01
- #print("back")
+ thisAgent.action.u[1] = self.allCheatPunishment # all cheat: move back ward
  elif numOfCheaters == 0:
- thisAgent.action.u[1] = 0.01 # all cooperate: move up together
+ thisAgent.action.u[1] = self.allCooperateForce # all cooperate: move up together
  else:
  if thisAgent.action.u[1] > 0.0: # this agent cheat
- thisAgent.action.u[1] = 0.02
- #print("cheat")
+ thisAgent.action.u[1] = self.cheaterGain
  else:
- thisAgent.action.u[1] = 0.01
+ thisAgent.action.u[1] = self.robbedCooperator
  reward_n.append(thisAgent.action.u[1])
  return reward_n
 
@@ -132,5 +112,4 @@ def observation(self, agent, world):
  entity_pos = []
  for entity in world.landmarks:
  entity_pos.append(entity.state.p_pos - agent.state.p_pos)
- return np.concatenate([agent.state.p_vel] + entity_pos)
-
+ return np.concatenate([agent.state.p_vel] + entity_pos)