default=False,
help='Agent Playing.')
args = parse.parse_args()
TEAM = 'HELIOS'
PORT = 6000
ACTOR_MODEL_NAME = "ppo_actor_go_to_ball"
CRITIC_MODEL_NAME = "ppo_critic_go_to_ball"
use_cuda = torch.cuda.is_available()
device = torch.device("cuda" if use_cuda else "cpu")
hfo_env = HFOEnv(is_offensive=True,
strict=True,
continuous=True,
team=TEAM,
port=PORT,
selected_action=DASH_ACTION,
selected_reward=GO_TO_BALL_REWARD,
selected_state=BALL_AXIS_POSITION_SPACE)
unum = hfo_env.getUnum()
params = PARAMS['ppo']
ppo = PPO(hfo_env.observation_space.shape[0], hfo_env.action_space.shape[0],
params)
def train():
writer = SummaryWriter('logs/{}_PPO_GO_TO_BALL'.format(
datetime.datetime.now().strftime("%Y-%m-%d_%H-%M-%S")))
Transition = namedtuple('Transition', ['s', 'a', 'a_log_p', 'r', 's_'])
try:
parse = argparse.ArgumentParser(
description='Agent Args', formatter_class=argparse.RawTextHelpFormatter)
parse.add_argument('--play', dest='play', action='store_true',
default=False, help='Agent Playing.')
args = parse.parse_args()
TEAM = 'HELIOS'
PORT = 6000
ACTOR_MODEL_NAME = "ddpg_actor_ball_to_goal"
CRITIC_MODEL_NAME = "ddpg_critic_ball_to_goal"
ENABLE_LOSS_WRITE = False
hfo_env = HFOEnv(is_offensive=True, strict=True,
continuous=True, team=TEAM, port=PORT,
selected_action=CONDITIONAL_DASH_OR_KICK_ACTION,
selected_reward=AGENT_AND_BALL_POTENCIAL_REWARD,
selected_state=AGENT_ORIENTATION_AND_BALL_POSITION_SPACE)
unum = hfo_env.getUnum()
params = PARAMS['ddpg']
ddpg = DDPG(
hfo_env.observation_space.shape[0], hfo_env.action_space.shape[0], params)
ou_noise = OUNoise(hfo_env.action_space)
replay_buffer = ReplayBuffer(params['replay_buffer_size'])
def train():
writer = SummaryWriter(
'logs/{}_DDPG_BALL_TO_GOAL'.format(datetime.datetime.now().strftime("%Y-%m-%d_%H-%M-%S")))
frame_idx = 0
import datetime
import logging
import os
import pickle
import itertools
import hfo
import numpy as np
from src.lib.hfo_env import HFOEnv
from src.lib.utils.action_selector import TEST_ACTION
team = 'HELIOS'
port = 6000
hfo_env = HFOEnv(is_offensive=True,
strict=True,
continuous=True,
team=team,
port=port,
selected_action=TEST_ACTION)
for episode in itertools.count():
status = hfo.IN_GAME
done = True
state = hfo_env.reset()
while status == hfo.IN_GAME:
next_state, reward, done, status = hfo_env.step([-0.6])
if done:
break
if status == hfo.SERVER_DOWN:
hfo_env.act(hfo.QUIT)