def get_reward(self, old_bird: Bird, bird: Bird) -> Tuple[bool, int]:
reset_bird = False
if bird.get_top() <= self.height and bird.get_bottom() > 0:
if self.board_controller.is_top(bird):
reward = Reward.REWARD_STUCK
reset_bird = True
elif self.board_controller.is_bottom(
bird) or self.board_controller.is_pipe(bird):
self.loose = True
reward = Reward.REWARD_LOOSE
if len(self.board_controller.goals) != 0:
self.board_controller.goals.pop(0)
elif self.board_controller.is_checkpoint(bird):
self.win_streak += 1
self.checked = True
reward = Reward.REWARD_CHECKPOINT
self.board_controller.goals.pop(0)
else:
reward = Reward.REWARD_DEFAULT
else:
reset_bird = True
reward = Reward.REWARD_IMPOSSIBLE
penalty = 0
if not reset_bird:
old_distance = self.board_controller.distance_next_pipe(old_bird)
distance = self.board_controller.distance_next_pipe(bird)
if distance == -1:
penalty = 0
elif old_distance - distance < 0:
penalty = distance * int(Reward.REWARD_PENALTY)
else:
penalty = int(Reward.REWARD_CHECKPOINT)
return reset_bird, int(reward) + penalty
def bind_genomes(self):
for genome_id, genome in self.__genomes:
genome.fitness = 0
net = neat.nn.FeedForwardNetwork.create(genome, self.__config)
self.__nets.append(net)
self.__birds.append(Bird(100, 266))
self.__ge.append(genome)
def update_bird(self, bird: Bird, action: str) -> Tuple[Bird, int]:
self.checked = False
old_bird = copy.deepcopy(bird)
if action == UP:
bird.flap()
elif action == RELEASE:
bird.fall()
reset_bird, reward = self.get_reward(old_bird, bird)
if reset_bird:
if action == UP:
bird.reset_flap()
return bird, reward
def is_top(self, bird: Bird) -> bool:
return bird.get_top() >= self.board.height - self.texture_manager.texture["top"]["height"]
def is_bottom(self, bird: Bird) -> bool:
return bird.get_bottom() < self.texture_manager.texture["bottom"]["height"]
def update_policy(self, bird: Bird) -> None:
self.policy.update(
self.previous_bird.get_state(), bird.get_state(),
self.environment.board_controller.get_position(self.previous_bird),
self.environment.board_controller.get_position(bird),
self.last_action, self.reward)
def best_action(self, bird: Bird) -> str:
return self.policy.best_action(
bird.get_state(),
self.environment.board_controller.get_position(bird))
def reset(self, bird: Bird) -> None:
self.environment.reset()
self.previous_bird = bird.reset()
self.score = 0
self.reward = 0
def in_pipe(self, pipe: Pipe, bird: Bird) -> bool:
if pipe.position_x <= bird.get_max_x() and pipe.position_x + self.texture["pipe"]["width"] >= bird.get_min_x() \
and (pipe.bottom >= bird.get_bottom() or bird.get_top() >= pipe.top):
return True
return False
def in_checkpoint(self, pipe: Pipe, bird: Bird) -> bool:
if pipe.position_x + self.texture["pipe"]["width"] <= bird.get_min_x() \
and pipe.bottom < bird.get_bottom() and bird.get_top() < pipe.top:
return True
return False
import arcade
from resources.env import SCREEN_WIDTH, SCREEN_HEIGHT
from src.controllers.environment import Environment
from src.entities.bird import Bird
from src.learning_engine.agent import Agent
from src.views.flappy_window import FlappyWindow
if __name__ == "__main__":
environment = Environment(SCREEN_WIDTH, SCREEN_HEIGHT)
bird = Bird()
agent = Agent(environment, bird)
window = FlappyWindow(agent, bird)
window.setup()
arcade.run()