def demo_populate_scene(self):
# agents
self.agents.add(
Agent(agent_id=0,
screen=self.screen,
game=self,
agent_image=self.agent_image,
field=self.field_rect,
init_position=(3, 1),
init_direction=(1, 1),
max_speed=1.8,
waypoints=[self.waypoints['stop'], self.waypoints['start']]))
self.agents.add(
Agent(agent_id=1,
screen=self.screen,
game=self,
agent_image=self.agent_image,
field=self.field_rect,
init_position=(3, 5),
init_direction=(1, 1),
max_speed=1.8,
waypoints=[self.waypoints['start'], self.waypoints['stop']]))
self.agents.add(
Agent(agent_id=2,
screen=self.screen,
game=self,
agent_image=self.agent_image,
field=self.field_rect,
init_position=(2, 2),
init_direction=(1, 1),
max_speed=1.14,
waypoints=[self.waypoints['start'], self.waypoints['stop']]))
def __init__(self, width, height, n_agents):
self.width = width
self.height = height
self.n_agents = n_agents
self.agent1 = Agent()
self.agent2 = Agent()
self.target = Target()
self.grid = np.zeros([self.width, self.height])
self.set_agent_pos()
self.set_target_pos()
self.obs = []
def add_agents(self, agent_dict):
for agent in agent_dict:
dx, dy = float(agent['dx']), float(agent['dy'])
x, y = float(agent['x']), float(agent['y'])
num = int(agent['n'])
a_type = int(agent['type'])
# spawn an agent in a random direction and position(within dx, dy)
direction = (randint(-1, 1), randint(-1, 1))
position = random_position(x, y, dx, dy)
waypoints = [awp['id'] for awp in agent['addwaypoint']]
rd = 0.3
for _ in xrange(num):
self.agents.add(
Agent(agent_id=self._agent_count,
atype=a_type,
screen=self.screen,
game=self,
agent_image=self.agent_image,
field=self.field_rect,
init_position=position,
init_direction=direction,
max_speed=normalvariate(1.34, 0.26),
radius=rd,
waypoints=[self.waypoints[wp] for wp in waypoints]))
self._agent_count += 1
def __init__(self, width, height, n_agents):
self.width = width
self.height = height
self.n_agents = n_agents
self.agents = [Agent()
for _ in range(n_agents)] # initialize multiple agents
self.target = Target()
self.grid = np.zeros([self.width, self.height])
self.timestep = 0
self.set_agent_pos()
self.set_target_pos()
self.obs = []
class Gridworld():
def __init__(self, width, height, n_agents):
self.width = width
self.height = height
self.n_agents = n_agents
self.agent1 = Agent()
self.agent2 = Agent()
self.target = Target()
self.grid = np.zeros([self.width, self.height])
self.set_agent_pos()
self.set_target_pos()
self.obs = []
def set_agent_pos(self):
pos_x = self.agent1.pos_x
pos_y = self.agent1.pos_y
pos_x2 = self.agent2.pos_x
pos_y2 = self.agent2.pos_y
self.grid[pos_x][pos_y] = 1
self.grid[pos_x2][pos_y2] = 2
def set_target_pos(self, random_pos=False):
if random_pos:
pos_x = random.randint(0, self.width)
pos_y = random.randint(0, self.height)
self.target.pos_x = pos_x #check if this is correct
self.target.pos_y = pos_y #check if thi is correct
else:
pos_x = self.target.pos_x
pos_y = self.target.pos_y
self.grid[pos_x][pos_y] = 8
def step(self, action, action2):
done = False
# Individual rewards
reward = 0
reward_2 = 0
reward -= 1 # negative reward for each step
reward_2 -= 1
# generate next action
self.agent1.update_pos(action)
self.agent2.update_pos(action2)
# update positions in the grid world
self.grid_reset()
self.set_agent_pos()
# dist = math.sqrt((self.agent1.pos_x - self.target.pos_x) ** 2 + (self.agent1.pos_y - self.target.pos_y) ** 2)
# reward = reward - abs(0.1*dist)
done1 = False
done2 = False
# if agent 1 reaches the target
if self.agent1.pos_x == self.target.pos_x and self.agent1.pos_y == self.target.pos_y:
reward += 20
done1 = True
# if agent2 reaches the target
if self.agent2.pos_x == self.target.pos_x and self.agent2.pos_y == self.target.pos_y:
reward_2 += 20
done2 = True
if (done1 == True) | (done2 == True):
done = True
# create observation vector of agent position(s) and target position
observations = []
observations.append(self.agent1.pos_x) # agent1
observations.append(self.agent1.pos_y)
observations.append(self.agent2.pos_x) # agent2
observations.append(self.agent2.pos_y)
observations.append(self.target.pos_x)
observations.append(self.target.pos_y)
#self.target.move_random() # TODO: remove this for stationary target
self.set_target_pos(random_pos=False)
return observations, reward, reward_2, done
def reset(self):
self.agent1.reset()
self.agent2.reset(
self.agent1.pos_x, self.agent1.pos_y,
use_given_start=True) # true to use same starting location
self.target.reset()
observations = []
observations.append(self.agent1.pos_x) # agent1
observations.append(self.agent1.pos_y)
observations.append(self.agent2.pos_x) # agent2
observations.append(self.agent2.pos_y)
observations.append(self.target.pos_x)
observations.append(self.target.pos_y)
return observations
def grid_reset(self):
self.grid = np.zeros([self.width, self.height])
class Gridworld():
def __init__(self, width, height, n_agents):
self.width = width
self.height = height
self.n_agents = n_agents
self.agent1 = Agent()
self.target = Target()
self.grid = np.zeros([self.width, self.height])
self.timestep = 0
self.set_agent_pos()
self.set_target_pos()
self.obs = []
def set_agent_pos(self):
pos_x = self.agent1.pos_x
pos_y = self.agent1.pos_y
self.grid[pos_x][pos_y] = 1
def set_target_pos(self, random_pos=False):
if random_pos:
pos_x = random.randint(0, self.width)
pos_y = random.randint(0, self.height)
else:
pos_x = self.target.pos_x
pos_y = self.target.pos_y
self.grid[pos_x][pos_y] = 8
def step(self, action):
done = False
reward = 0
reward -= 1 # negative reward for each step
# generate next action
self.agent1.update_pos(action)
self.target.move_random() # TODO: remove this stationary target
# update positions in the grid world
self.grid_reset()
self.set_agent_pos()
self.set_target_pos(random_pos=False)
# dist = math.sqrt((self.agent1.pos_x - self.target.pos_x) ** 2 + (self.agent1.pos_y - self.target.pos_y) ** 2)
# reward = reward - abs(0.1*dist)
if self.agent1.pos_x == self.target.pos_x and self.agent1.pos_y == self.target.pos_y:
reward += 20
done = True
self.timestep += 1
#print("reward : "+str(reward)+str("\n"))
# create observation vector of agent position(s) and target position
observations = []
observations.append(self.agent1.pos_x)
observations.append(self.agent1.pos_y)
observations.append(self.target.pos_x)
observations.append(self.target.pos_y)
return observations, reward, done
def reset(self):
self.timestep = 0
self.agent1.reset()
self.target.reset()
observations = []
observations.append(self.agent1.pos_x)
observations.append(self.agent1.pos_y)
observations.append(self.target.pos_x)
observations.append(self.target.pos_y)
return observations
def grid_reset(self):
self.grid = np.zeros([self.width, self.height])