def main():
param_noise = None
env1 = tm700GymEnv2(renders=False, isDiscrete=False)
model = DDPG(MlpPolicy,
env1,
verbose=1,
param_noise=param_noise,
random_exploration=0.1)
# model = DQN(MlpPolicy, env1, verbose=1, exploration_fraction=0.3)
# = deepq.models.mlp([64])
start = time.time()
model.learn(total_timesteps=1000000)
#max_timesteps=10000000,
# exploration_fraction=0.1,
# exploration_final_eps=0.02,
# print_freq=10,
# callback=callback, network='mlp')
print("Saving model")
model.save("tm_test_model_randomblocksrotated.pkl")
print('total time', time.time() - start)
import os, inspect
currentdir = os.path.dirname(os.path.abspath(inspect.getfile(inspect.currentframe())))
parentdir = os.path.dirname(os.path.dirname(currentdir))
os.sys.path.insert(0, parentdir)
import gym
from pybullet_envs.bullet.tm700GymEnv_TEST import tm700GymEnv2
from stable_baselines import DQN, PPO2, DDPG
from pybullet_envs.baselines.train_tm700_multivec import evaluate, record_video
#################### PARAMETERS
savedmodel = "tm_test_model_randomblocks.pkl"
env = tm700GymEnv2(renders=True, isDiscrete=False)
model = DDPG.load(savedmodel, env=env)
########## run simulation
def runsimulation(model, env, iterations):
obs = env.reset()
time_step_counter = 0
iterations = iterations
while time_step_counter < iterations:
action, _ = model.predict(obs)
obs, rewards, dones, _ = env.step(action) # Assumption: eval conducted on single env only!
time_step_counter +=1
# time.sleep(0.1)
if dones:
"""
Utility function for multiprocessed env.
:param env_id: (str) the environment ID
:param seed: (int) the inital seed for RNG
:param rank: (int) index of the subprocess
"""
def _init():
env = gym.make(env_id)
# Important: use a different seed for each environment
env.seed(seed + rank)
return env
set_global_seeds(seed)
return _init
if __name__ == '__main__':
env_id = tm700GymEnv2(renders=True, isDiscrete=True)
# The different number of processes that will be used
PROCESSES_TO_TEST = [1, 2, 4, 8, 16]
NUM_EXPERIMENTS = 3 # RL algorithms can often be unstable, so we run several experiments (see https://arxiv.org/abs/1709.06560)
TRAIN_STEPS = 5000
# Number of episodes for evaluation
EVAL_EPS = 20
ALGO = PPO2
# We will create one environment to evaluate the agent on
eval_env = DummyVecEnv([lambda: gym.make(env_id)])
# from stable_baselines.common.env_checker import check_env
from pybullet_envs.bullet.tm700GymEnv_TEST import tm700GymEnv2
env = tm700GymEnv2()
# It will check your custom environment and output additional warnings if needed
# check_env(env)
obs = env.reset()
n_steps = 10
for _ in range(n_steps):
# Random action
action = env.action_space.sample()
obs, reward, done, info = env.step(action)