gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2025-12-17 11:10:20 +01:00
Code Issues Projects Releases Wiki Activity

Release 0.9

Browse Source 
Main changes are detailed below:

New features -
* CARLA 0.7 simulator integration
* Human control of the game play
* Recording of human game play and storing / loading the replay buffer
* Behavioral cloning agent and presets
* Golden tests for several presets
* Selecting between deep / shallow image embedders
* Rendering through pygame (with some boost in performance)

API changes -
* Improved environment wrapper API
* Added an evaluate flag to allow convenient evaluation of existing checkpoints
* Improve frameskip definition in Gym

Bug fixes -
* Fixed loading of checkpoints for agents with more than one network
* Fixed the N Step Q learning agent python3 compatibility
This commit is contained in:
Itai Caspi
2017-12-19 19:27:16 +02:00
committed by GitHub
parent 11faf19649
commit 125c7ee38d
41 changed files with 1713 additions and 260 deletions
Download Patch File Download Diff File Expand all files Collapse all files

90
environments/doom_environment_wrapper.py
View File

@@ -25,6 +25,7 @@ import numpy as np
from environments.environment_wrapper import EnvironmentWrapper
from os import path, environ
from utils import *
from logger import *
# enum of the available levels and their path
@@ -39,6 +40,43 @@ class DoomLevel(Enum):
DEFEND_THE_LINE = "defend_the_line.cfg"
DEADLY_CORRIDOR = "deadly_corridor.cfg"
key_map = {
'NO-OP': 96, # `
'ATTACK': 13, # enter
'CROUCH': 306, # ctrl
'DROP_SELECTED_ITEM': ord("t"),
'DROP_SELECTED_WEAPON': ord("t"),
'JUMP': 32, # spacebar
'LAND': ord("l"),
'LOOK_DOWN': 274, # down arrow
'LOOK_UP': 273, # up arrow
'MOVE_BACKWARD': ord("s"),
'MOVE_DOWN': ord("s"),
'MOVE_FORWARD': ord("w"),
'MOVE_LEFT': 276,
'MOVE_RIGHT': 275,
'MOVE_UP': ord("w"),
'RELOAD': ord("r"),
'SELECT_NEXT_WEAPON': ord("q"),
'SELECT_PREV_WEAPON': ord("e"),
'SELECT_WEAPON0': ord("0"),
'SELECT_WEAPON1': ord("1"),
'SELECT_WEAPON2': ord("2"),
'SELECT_WEAPON3': ord("3"),
'SELECT_WEAPON4': ord("4"),
'SELECT_WEAPON5': ord("5"),
'SELECT_WEAPON6': ord("6"),
'SELECT_WEAPON7': ord("7"),
'SELECT_WEAPON8': ord("8"),
'SELECT_WEAPON9': ord("9"),
'SPEED': 304, # shift
'STRAFE': 9, # tab
'TURN180': ord("u"),
'TURN_LEFT': ord("a"), # left arrow
'TURN_RIGHT': ord("d"), # right arrow
'USE': ord("f"),
}
class DoomEnvironmentWrapper(EnvironmentWrapper):
def __init__(self, tuning_parameters):
@@ -49,26 +87,42 @@ class DoomEnvironmentWrapper(EnvironmentWrapper):
self.scenarios_dir = path.join(environ.get('VIZDOOM_ROOT'), 'scenarios')
self.game = vizdoom.DoomGame()
self.game.load_config(path.join(self.scenarios_dir, self.level))
self.game.set_window_visible(self.is_rendered)
self.game.set_window_visible(False)
self.game.add_game_args("+vid_forcesurface 1")
if self.is_rendered:
self.wait_for_explicit_human_action = True
if self.human_control:
self.game.set_screen_resolution(vizdoom.ScreenResolution.RES_640X480)
self.renderer.create_screen(640, 480)
elif self.is_rendered:
self.game.set_screen_resolution(vizdoom.ScreenResolution.RES_320X240)
self.renderer.create_screen(320, 240)
else:
# lower resolution since we actually take only 76x60 and we don't need to render
self.game.set_screen_resolution(vizdoom.ScreenResolution.RES_160X120)
self.game.set_render_hud(False)
self.game.set_render_crosshair(False)
self.game.set_render_decals(False)
self.game.set_render_particles(False)
self.game.init()
# action space
self.action_space_abs_range = 0
self.actions = {}
self.action_space_size = self.game.get_available_buttons_size()
for action_idx in range(self.action_space_size):
self.actions[action_idx] = [0] * self.action_space_size
self.actions[action_idx][action_idx] = 1
self.actions_description = [str(action) for action in self.game.get_available_buttons()]
self.action_space_size = self.game.get_available_buttons_size() + 1
self.action_vector_size = self.action_space_size - 1
self.actions[0] = [0] * self.action_vector_size
for action_idx in range(self.action_vector_size):
self.actions[action_idx + 1] = [0] * self.action_vector_size
self.actions[action_idx + 1][action_idx] = 1
self.actions_description = ['NO-OP']
self.actions_description += [str(action).split(".")[1] for action in self.game.get_available_buttons()]
for idx, action in enumerate(self.actions_description):
if action in key_map.keys():
self.key_to_action[(key_map[action],)] = idx
# measurement
self.measurements_size = self.game.get_state().game_variables.shape
self.width = self.game.get_screen_width()
@@ -77,27 +131,17 @@ class DoomEnvironmentWrapper(EnvironmentWrapper):
self.game.set_seed(self.tp.seed)
self.reset()
def _update_observation_and_measurements(self):
def _update_state(self):
# extract all data from the current state
state = self.game.get_state()
if state is not None and state.screen_buffer is not None:
self.observation = self._preprocess_observation(state.screen_buffer)
self.observation = state.screen_buffer
self.measurements = state.game_variables
self.reward = self.game.get_last_reward()
self.done = self.game.is_episode_finished()
def step(self, action_idx):
self.reward = 0
for frame in range(self.tp.env.frame_skip):
self.reward += self.game.make_action(self._idx_to_action(action_idx))
self._update_observation_and_measurements()
if self.done:
break
return {'observation': self.observation,
'reward': self.reward,
'done': self.done,
'action': action_idx,
'measurements': self.measurements}
def _take_action(self, action_idx):
self.game.make_action(self._idx_to_action(action_idx), self.frame_skip)
def _preprocess_observation(self, observation):
if observation is None:
@@ -108,3 +152,5 @@ class DoomEnvironmentWrapper(EnvironmentWrapper):
def _restart_environment_episode(self, force_environment_reset=False):
self.game.new_episode()