epochs = rospy.get_param("/ML/epochs")
hdim = rospy.get_param("/ML/hdim")
policy_lr = rospy.get_param("/ML/policy_lr")
value_lr = rospy.get_param("/ML/value_lr")
max_std = rospy.get_param("/ML/max_std")
clip_range = rospy.get_param("/ML/clip_range")
n_step = rospy.get_param("/ML/n_step")
gamma = rospy.get_param("/ML/gamma")
lam = rospy.get_param("/ML/lam")
episode_size = rospy.get_param("/ML/episode_size")
batch_size = rospy.get_param("/ML/batch_size")
nupdates = rospy.get_param("/ML/nupdates")
maxlen_num = rospy.get_param("/ML/maxlen_num")
agent = PPOGAEAgent(obs_dim, n_act, epochs, hdim, policy_lr, value_lr, max_std, clip_range, seed)
#agent = PPOGAEAgent(obs_dim, n_act, epochs=10, hdim=obs_dim, policy_lr=3e-3, value_lr=1e-3, max_std=1.0, clip_range=0.2, seed=seed)
'''
PPO Agent with Gaussian policy
'''
def run_episode(env, animate=False): # Run policy and collect (state, action, reward) pairs
obs = env.reset()
observes, actions, rewards, infos = [], [], [], []
done = False
for update in range(n_step):
print("update", update)
obs = np.array(obs)
obs = obs.astype(np.float32).reshape((1, -1)) # numpy.ndarray (1, num_obs)
observes.append(obs)
# import our training environment
import gym
from env.ur_door_opening_env import URSimDoorOpening
# import our training algorithms
from algorithm.ppo_gae import PPOGAEAgent
seed = 0
obs_dim = 21 # env.observation_space.shape[0] # have to change number of hdim
n_act = 6 #config: act_dim #env.action_space.n
agent = PPOGAEAgent(obs_dim,
n_act,
epochs=10,
hdim=64,
policy_lr=1e-4,
value_lr=1e-4,
max_std=1.0,
clip_range=0.2,
seed=seed)
#agent = PPOGAEAgent(obs_dim, n_act, epochs=10, hdim=obs_dim, policy_lr=3e-3, value_lr=1e-3, max_std=1.0, clip_range=0.2, seed=seed)
'''
PPO Agent with Gaussian policy
'''
def run_episode(
env,
animate=False): # Run policy and collect (state, action, reward) pairs
obs = env.reset()
observes, actions, rewards, infos = [], [], [], []