File size: 14,135 Bytes
079c32c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 |
"""
The code below is adapted from https://github.com/lich14/CDS/tree/main/CDS_GRF/envs/grf,
which is from the codebase of the CDS paper "Celebrating Diversity in Shared Multi-Agent Reinforcement Learning"
"""
import gfootball.env as football_env
from gfootball.env import observation_preprocessing
import gym
import numpy as np
from ding.utils import ENV_REGISTRY
from typing import Any, List, Union, Optional
import copy
import torch
from ding.envs import BaseEnv, BaseEnvTimestep
from ding.torch_utils import to_ndarray, to_list
import os
from matplotlib import animation
import matplotlib.pyplot as plt
@ENV_REGISTRY.register('gfootball-academy')
class GfootballAcademyEnv(BaseEnv):
def __init__(
self,
cfg: dict,
dense_reward=False,
write_full_episode_dumps=False,
write_goal_dumps=False,
dump_freq=1000,
render=False,
time_limit=150,
time_step=0,
stacked=False,
representation="simple115",
rewards='scoring',
logdir='football_dumps',
write_video=True,
number_of_right_players_agent_controls=0,
):
"""
'academy_3_vs_1_with_keeper'
n_agents=3,
obs_dim=26,
'academy_counterattack_hard'
n_agents=4,
obs_dim=34,
"""
self._cfg = cfg
self._save_replay = False
self._save_replay_count = 0
self._replay_path = None
self.dense_reward = dense_reward
self.write_full_episode_dumps = write_full_episode_dumps
self.write_goal_dumps = write_goal_dumps
self.dump_freq = dump_freq
self.render = render
self.env_name = self._cfg.env_name # TODO
self.n_agents = self._cfg.agent_num
self.obs_dim = self._cfg.obs_dim
self.episode_limit = time_limit
self.time_step = time_step
self.stacked = stacked
self.representation = representation
self.rewards = rewards
self.logdir = logdir
self.write_video = write_video
self.number_of_right_players_agent_controls = number_of_right_players_agent_controls
self._env = football_env.create_environment(
write_full_episode_dumps=self.write_full_episode_dumps,
write_goal_dumps=self.write_goal_dumps,
env_name=self.env_name,
stacked=self.stacked,
representation=self.representation,
rewards=self.rewards,
logdir=self.logdir,
render=self.render,
write_video=self.write_video,
dump_frequency=self.dump_freq,
number_of_left_players_agent_controls=self.n_agents,
number_of_right_players_agent_controls=self.number_of_right_players_agent_controls,
channel_dimensions=(observation_preprocessing.SMM_WIDTH, observation_preprocessing.SMM_HEIGHT)
)
obs_space_low = self._env.observation_space.low[0][:self.obs_dim]
obs_space_high = self._env.observation_space.high[0][:self.obs_dim]
self._action_space = gym.spaces.Dict(
{agent_i: gym.spaces.Discrete(self._env.action_space.nvec[1])
for agent_i in range(self.n_agents)}
)
self._observation_space = gym.spaces.Dict(
{
agent_i:
gym.spaces.Box(low=obs_space_low, high=obs_space_high, dtype=self._env.observation_space.dtype)
for agent_i in range(self.n_agents)
}
)
self._reward_space = gym.spaces.Box(low=0, high=100, shape=(1, ), dtype=np.float32) # TODO(pu)
self.n_actions = self.action_space[0].n
def get_simple_obs(self, index=-1):
full_obs = self._env.unwrapped.observation()[0]
simple_obs = []
if self.env_name == 'academy_3_vs_1_with_keeper':
if index == -1:
# global state, absolute position
simple_obs.append(full_obs['left_team'][-self.n_agents:].reshape(-1))
simple_obs.append(full_obs['left_team_direction'][-self.n_agents:].reshape(-1))
simple_obs.append(full_obs['right_team'].reshape(-1))
simple_obs.append(full_obs['right_team_direction'].reshape(-1))
simple_obs.append(full_obs['ball'])
simple_obs.append(full_obs['ball_direction'])
else:
# local state, relative position
ego_position = full_obs['left_team'][-self.n_agents + index].reshape(-1)
simple_obs.append(ego_position)
simple_obs.append(
(np.delete(full_obs['left_team'][-self.n_agents:], index, axis=0) - ego_position).reshape(-1)
)
simple_obs.append(full_obs['left_team_direction'][-self.n_agents + index].reshape(-1))
simple_obs.append(
np.delete(full_obs['left_team_direction'][-self.n_agents:], index, axis=0).reshape(-1)
)
simple_obs.append((full_obs['right_team'] - ego_position).reshape(-1))
simple_obs.append(full_obs['right_team_direction'].reshape(-1))
simple_obs.append(full_obs['ball'][:2] - ego_position)
simple_obs.append(full_obs['ball'][-1].reshape(-1))
simple_obs.append(full_obs['ball_direction'])
elif self.env_name == 'academy_counterattack_hard':
if index == -1:
# global state, absolute position
simple_obs.append(full_obs['left_team'][-self.n_agents:].reshape(-1))
simple_obs.append(full_obs['left_team_direction'][-self.n_agents:].reshape(-1))
simple_obs.append(full_obs['right_team'][0])
simple_obs.append(full_obs['right_team'][1])
simple_obs.append(full_obs['right_team'][2])
simple_obs.append(full_obs['right_team_direction'][0])
simple_obs.append(full_obs['right_team_direction'][1])
simple_obs.append(full_obs['right_team_direction'][2])
simple_obs.append(full_obs['ball'])
simple_obs.append(full_obs['ball_direction'])
else:
# local state, relative position
ego_position = full_obs['left_team'][-self.n_agents + index].reshape(-1)
simple_obs.append(ego_position)
simple_obs.append(
(np.delete(full_obs['left_team'][-self.n_agents:], index, axis=0) - ego_position).reshape(-1)
)
simple_obs.append(full_obs['left_team_direction'][-self.n_agents + index].reshape(-1))
simple_obs.append(
np.delete(full_obs['left_team_direction'][-self.n_agents:], index, axis=0).reshape(-1)
)
simple_obs.append(full_obs['right_team'][0] - ego_position)
simple_obs.append(full_obs['right_team'][1] - ego_position)
simple_obs.append(full_obs['right_team'][2] - ego_position)
simple_obs.append(full_obs['right_team_direction'][0])
simple_obs.append(full_obs['right_team_direction'][1])
simple_obs.append(full_obs['right_team_direction'][2])
simple_obs.append(full_obs['ball'][:2] - ego_position)
simple_obs.append(full_obs['ball'][-1].reshape(-1))
simple_obs.append(full_obs['ball_direction'])
simple_obs = np.concatenate(simple_obs)
return simple_obs
def get_global_state(self):
return self.get_simple_obs(-1)
def get_global_special_state(self):
return [np.concatenate([self.get_global_state(), self.get_obs_agent(i)]) for i in range(self.n_agents)]
def check_if_done(self):
cur_obs = self._env.unwrapped.observation()[0]
ball_loc = cur_obs['ball']
ours_loc = cur_obs['left_team'][-self.n_agents:]
if ball_loc[0] < 0 or any(ours_loc[:, 0] < 0):
"""
This is based on the CDS paper:
'We make a small and reasonable change to the half-court offensive scenarios: our players will lose if
they or the ball returns to our half-court.'
"""
return True
return False
def reset(self):
"""Returns initial observations and states."""
if self._save_replay:
self._frames = []
self.time_step = 0
self._env.reset()
obs = {
'agent_state': np.stack(self.get_obs(), axis=0).astype(np.float32),
# Note: here 'global_state' is the agent_specific_global_state,
# we simply concatenate the global_state and agent_state
'global_state': np.stack(
self.get_global_special_state(),
axis=0,
).astype(np.float32),
'action_mask': np.stack(self.get_avail_actions(), axis=0).astype(np.float32),
}
if hasattr(self, '_seed') and hasattr(self, '_dynamic_seed') and self._dynamic_seed:
np_seed = 100 * np.random.randint(1, 1000)
self._env.seed(self._seed + np_seed)
elif hasattr(self, '_seed'):
self._env.seed(self._seed)
self._eval_episode_return = 0
return obs
def step(self, actions):
"""Returns reward, terminated, info."""
assert isinstance(actions, np.ndarray) or isinstance(actions, list), type(actions)
self.time_step += 1
if isinstance(actions, np.ndarray):
actions = actions.tolist()
if self._save_replay:
self._frames.append(self._env.render(mode='rgb_array'))
_, original_rewards, done, infos = self._env.step(actions)
obs = {
'agent_state': np.stack(self.get_obs(), axis=0).astype(np.float32),
# Note: here 'global_state' is the agent_specific_global_state,
# we simply concatenate the global_state and agent_state
'global_state': np.stack(
self.get_global_special_state(),
axis=0,
).astype(np.float32),
'action_mask': np.stack(self.get_avail_actions(), axis=0).astype(np.float32),
}
rewards = list(original_rewards)
if self.time_step >= self.episode_limit:
done = True
if self.check_if_done():
done = True
if done:
if self._save_replay:
path = os.path.join(
self._replay_path, '{}_episode_{}.gif'.format(self.env_name, self._save_replay_count)
)
self.display_frames_as_gif(self._frames, path)
self._save_replay_count += 1
if sum(rewards) <= 0:
"""
This is based on the CDS paper:
"Environmental reward only occurs at the end of the game.
They will get +100 if they win, else get -1."
If done=False, the reward is -1,
If done=True and sum(rewards)<=0 the reward is 1.
If done=True and sum(rewards)>0 the reward is 100.
"""
infos['eval_episode_return'] = infos['score_reward'] # TODO(pu)
return BaseEnvTimestep(obs, np.array(-int(done)).astype(np.float32), done, infos)
else:
infos['eval_episode_return'] = infos['score_reward']
return BaseEnvTimestep(obs, np.array(100).astype(np.float32), done, infos)
def get_obs(self):
"""Returns all agent observations in a list."""
obs = [self.get_simple_obs(i) for i in range(self.n_agents)]
return obs
def get_obs_agent(self, agent_id):
"""Returns observation for agent_id."""
return self.get_simple_obs(agent_id)
def get_obs_size(self):
"""Returns the size of the observation."""
return self.obs_dim
def get_state(self):
"""Returns the global state."""
return self.get_global_state()
def get_state_size(self):
"""Returns the size of the global state."""
return self.obs_dim
def get_avail_actions(self):
"""Returns the available actions of all agents in a list."""
return [[1 for _ in range(self.n_actions)] for agent_id in range(self.n_agents)]
def get_avail_agent_actions(self, agent_id):
"""Returns the available actions for agent_id."""
return self.get_avail_actions()[agent_id]
def render(self):
pass
def close(self):
self._env.close()
def save_replay(self):
"""Save a replay."""
pass
def seed(self, seed: int, dynamic_seed: bool = True) -> None:
self._seed = seed
self._dynamic_seed = dynamic_seed
np.random.seed(self._seed)
def random_action(self) -> np.ndarray:
random_action = self.action_space.sample()
random_action = to_ndarray([random_action], dtype=np.int64)
return random_action
@property
def observation_space(self) -> gym.spaces.Space:
return self._observation_space
@property
def action_space(self) -> gym.spaces.Space:
return self._action_space
@property
def reward_space(self) -> gym.spaces.Space:
return self._reward_space
def __repr__(self) -> str:
return f'GfootballEnv Academy Env {self.env_name}'
def enable_save_replay(self, replay_path: Optional[str] = None) -> None:
"""
Overview:
Save replay file in the given path
Arguments:
- replay_path(:obj:`str`): Storage path.
"""
if replay_path is None:
replay_path = './video'
self._save_replay = True
self._replay_path = replay_path
self._save_replay_count = 0
@staticmethod
def display_frames_as_gif(frames: list, path: str) -> None:
patch = plt.imshow(frames[0])
plt.axis('off')
def animate(i):
patch.set_data(frames[i])
anim = animation.FuncAnimation(plt.gcf(), animate, frames=len(frames), interval=5)
anim.save(path, writer='imagemagick', fps=20)
|