def __init__(self):
self.game = PyGameObjectAvoidance()
self.state = None
# Boundaries of observation space
self.low = np.array([0, 0, 0], dtype=np.int)
self.high = np.array([4, 4, 4], dtype=np.int)
# Defining the observation and action spaces
self.action_space = spaces.Discrete(3)
self.observation_space = spaces.Box(self.low, self.high, dtype=np.int)
def reset(self):
del self.game
self.game = PyGameObjectAvoidance()
# Makes a list of obstacles, 200-300pixels away from one another
newpos = 0
for i in range(50):
self.game.debris.append(Debris(window_width + newpos))
rand = random.randint(0, 2)
newpos += np.random.randint(200, 350)
# Updates the obstacles so they can move
for i in self.game.debris:
i.update()
# Returns the initial state of the environment, 3-tuple of the positions of the satellite and the 2 closest obstacles
self.state = (self.game.satellite.position_in_quantised_space(),
self.game.debris[0].position_in_quantised_space(),
self.game.debris[1].position_in_quantised_space())
return np.array(self.state)
class SatelliteEnv(gym.Env):
metadata = {
'render.modes': ['human', 'rgb_array'],
'video.frames_per_second': 30
}
def __init__(self):
self.game = PyGameObjectAvoidance()
self.state_satellite = None
self.state_debris = None
# Boundaries of observation spaces
self.low_satellite = np.array([0, 0, 0], dtype=np.int)
self.high_satellite = np.array([4, 4, 4], dtype=np.int)
self.low_debris = np.array([0, 0], dtype=np.int)
self.high_debris = np.array([4, 4], dtype=np.int)
# Defining the observation and action spaces
self.action_space_satellite = spaces.Discrete(3)
self.action_space_debris = spaces.Discrete(3)
self.observation_space_satellite = spaces.Box(self.low_satellite,
self.high_satellite,
dtype=np.int)
self.observation_space_debris = spaces.Box(self.low_debris,
self.high_debris,
dtype=np.int)
# Resets the environment
def reset(self):
del self.game
self.game = PyGameObjectAvoidance()
# Makes a list of obstacles, 200-300pixels away from one another
newpos = 0
for i in range(50):
rand = random.randint(0, 2)
self.game.debris.append(Debris(window_width + newpos))
newpos += np.random.randint(270, 310)
# Updates the obstacles so they can move
for i in self.game.debris:
i.update()
# Returns the initial state of the environment, for both agents
self.state_satellite = (
self.game.satellite.position_in_quantised_space(),
self.game.debris[0].position_in_quantised_space(),
self.game.debris[1].position_in_quantised_space())
self.state_debris = (self.game.satellite.position_in_quantised_space(),
self.game.debris[0].position_in_quantised_space())
return np.array(self.state_satellite), np.array(self.state_debris)
# One-step function of the environment
def step(self, actions):
action = actions[0]
action_d = actions[1]
assert self.action_space_satellite.contains(
action), "%r (%s) invalid" % (action, type(action))
assert self.action_space_debris.contains(
action_d), "%r (%s) invalid" % (action_d, type(action_d))
# The agents observes current state
obs_satellite = self.state_satellite
obs_debris = self.state_debris
reward_satellite = 0
reward_debris = 0
done = False
# The agents takes the action
self.game.satellite.action(action)
self.game.satellite.took_action = True
i = self.game.debris[0]
if i.posx < 80 and self.game.debris[1].posx < 200:
self.game.debris[1].action(action_d)
self.game.debris[0].action(action_d)
elif i.posx < 200:
i.action(action_d)
# Observe state at t+1
obs_satellite = (self.game.satellite.position_in_quantised_space(),
self.game.debris[0].position_in_quantised_space(),
self.game.debris[1].position_in_quantised_space())
obs_debris = (self.game.satellite.position_in_quantised_space(),
self.game.debris[0].position_in_quantised_space())
# For the nearest obstacle, check if it is out of the environment
if i.out_of_window:
self.game.debris.pop(self.game.debris.index(i))
# Determine the reward
if i.had_collision(self.game.satellite):
reward_satellite = -100
reward_debris = 100
self.game.satellite.is_alive = False
elif self.game.satellite.not_collision(i):
reward_satellite = 10
reward_debris = -10
elif self.game.satellite.satellite_aligned_w_obstacle(i):
reward_satellite = -1
reward_debris = 1
# Updates scoreboard
if self.game.satellite.is_alive and i.out_of_window:
if not i.checked:
self.game.satellite.objects_avoided += 1
i.checked = True
# If satellite is dead the environment needs to reset
if not self.game.satellite.is_alive:
done = True
obs = obs_satellite, obs_debris
reward = reward_satellite, reward_debris
return obs, reward, done, {}
# Render the environment
def render(self, mode="human", close=False):
self.game.view()
class SatelliteEnv(gym.Env):
metadata = {
'render.modes': ['human', 'rgb_array'],
'video.frames_per_second': 30
}
def __init__(self):
self.game = PyGameObjectAvoidance()
self.state = None
# Boundaries of observation space
self.low = np.array([0, 0, 0, 0, 0], dtype=np.int)
self.high = np.array([4, 4, 4, 4, 4], dtype=np.int)
# Defining the observation and action spaces
self.action_space = spaces.Discrete(3)
self.observation_space = spaces.Box(self.low, self.high, dtype=np.int)
# self.flag_move_deb = None
self.newpos = None
# Resets the environment
def reset(self):
del self.game
self.game = PyGameObjectAvoidance()
# Makes a list of obstacles, 200-300pixels away from one another
self.newpos = 0
# self.flag_move_deb = 0
for i in range(50):
rand = random.randint(0, 2)
self.game.debris.append(Debris(window_width + self.newpos))
self.newpos += np.random.randint(200, 350)
# Updates the obstacles so they can move
for i in self.game.debris:
i.update()
# Returns the initial state of the environment, 3-tuple of the positions of the satellite and the 2 closest obstacles
self.state = (self.game.satellite.position_in_quantised_space(),
self.game.debris[0].position_in_quantised_space()[0],
self.game.debris[0].position_in_quantised_space()[1],
self.game.debris[1].position_in_quantised_space()[0],
self.game.debris[1].position_in_quantised_space()[1])
return np.array(self.state)
# One-step function of the environment
def step(self, action):
# Implements movement of the debris as the DQN network was having issues
for j in self.game.debris:
if (j.posx > -2):
j.posx -= 4
else:
j.out_of_window = True
self.game.debris.append(Debris(window_width + self.newpos))
self.newpos += np.random.randint(200, 350)
assert self.action_space.contains(
action), "%r (%s) invalid" % (action, type(action))
# The agent observes current state
obs = self.state
reward = 0
done = False
# The agent takes the action
self.game.satellite.action(action)
# Observe state at t+1
obs = (self.game.satellite.position_in_quantised_space(),
self.game.debris[0].position_in_quantised_space()[0],
self.game.debris[0].position_in_quantised_space()[1],
self.game.debris[1].position_in_quantised_space()[0],
self.game.debris[1].position_in_quantised_space()[1])
# For the nearest obstacle, check if it is out of the environment
i = self.game.debris[0]
if i.out_of_window:
self.game.debris.pop(self.game.debris.index(i))
# Determine the reward
if i.had_collision(self.game.satellite):
reward = -100
self.game.satellite.is_alive = False
elif self.game.satellite.not_collision(i):
reward = 10
elif self.game.satellite.satellite_aligned_w_obstacle(i):
reward = -1
# Updates scoreboard
if self.game.satellite.is_alive and i.out_of_window:
if not i.checked:
self.game.satellite.objects_avoided += 1
i.checked = True
# If satellite is dead the environment needs to reset
if not self.game.satellite.is_alive:
done = True
return obs, reward, done, {}
# Render the environment
def render(self, mode="human", close=False):
self.game.view()