def test_simple_ltl_env(env, sampler):
env = gym.make(env)
env = ltl_wrappers.LTLEnv(env,
progression_mode="full",
ltl_sampler=sampler)
letter_types = env.propositions
def valid_action(a):
return a in letter_types
for _ in range(10):
obs = env.reset()
for _ in range(10000):
env.show()
print(obs["text"])
print("\nAction? ", end="")
a = input().strip()
while not valid_action(a):
a = input().strip()
a = letter_types.index(a)
obs, reward, done, info = env.step(a)
if done:
env.show()
print(reward)
print("Done!")
input()
break
print(reward)
env.close()
def test_env(env, sampler):
env = gym.make(env)
#env = ltl_wrappers.LTLEnv(env, progression_mode="full", ltl_sampler="Sequence_2_3")
env = ltl_wrappers.LTLEnv(env,
progression_mode="full",
ltl_sampler=sampler)
str_to_action = {"w": 0, "s": 1, "a": 2, "d": 3}
import random
for _ in range(10):
obs = env.reset()
for _ in range(10000):
env.show()
print(obs["text"])
print("\nAction? ", end="")
a = input()
while a not in str_to_action:
a = input()
print()
a = str_to_action[a]
#a = random.randrange(env.action_space.n)
obs, reward, done, info = env.step(a % env.action_space.n)
if done:
env.show()
print(reward)
print("Done!")
input()
break
print(reward)
env.close()
def make_env(env_key,
progression_mode,
ltl_sampler,
seed=None,
intrinsic=0,
noLTL=False):
env = gym.make(env_key)
env.seed(seed)
# Adding LTL wrappers
if (noLTL):
return ltl_wrappers.NoLTLWrapper(env)
else:
return ltl_wrappers.LTLEnv(env, progression_mode, ltl_sampler,
intrinsic)
def make_env(env_id,
mpi_rank=0,
subrank=0,
seed=None,
logger_dir=None,
initializer=None):
if initializer is not None:
initializer(mpi_rank=mpi_rank, subrank=subrank)
env = gym.make(env_id)
# Adding general wraps
env = ltl_wrappers.LTLEnv(env)
env = ltl_wrappers.NoLTLWrapper(env) # For testing purposes
env.seed(seed + subrank if seed is not None else None)
env = bench.Monitor(env,
logger_dir
and os.path.join(logger_dir,
str(mpi_rank) + '.' + str(subrank)),
allow_early_resets=True)
return env
type=int,
help="random seed to generate the environment with",
default=-1)
parser.add_argument("--tile_size",
type=int,
help="size at which to render tiles",
default=32)
parser.add_argument('--agent_view',
default=False,
help="draw the agent sees (partially observable view)",
action='store_true')
args = parser.parse_args()
# `base_env` is the backend minigrid
# `env` is the (1-level) wrapped minigrid from our code
# `ltl_env` is the (2-level) wrapped minigrid with LTL goals
env = gym.make(args.env)
base_env = env.env
ltl_env = ltl_wrappers.LTLEnv(env,
progression_mode="full",
ltl_sampler="AdversarialSampler")
window = Window('gym_minigrid - ' + args.env)
window.reg_key_handler(key_handler)
reset()
# Blocking event loop
window.show(block=True)
help='The path to the model to load.')
parser_viz.add_argument(
"--ltl-sampler",
default="Default",
help="the ltl formula template to sample from (default: DefaultSampler)"
)
args = vars(parser.parse_args()) # make it a dictionary
outdir = './storage/random-agent-results'
if (args["command"] == "play"):
env = gym.make(args["env_id"])
env.num_steps = 10000000
env = safety_wrappers.Play(env)
env = ltl_wrappers.LTLEnv(env, ltl_sampler="Default")
agent = PlayAgent(env)
elif (args["command"] == "random"):
env = gym.make(args["env_id"])
env.num_steps = 10000
env = safety_wrappers.Play(env)
env = ltl_wrappers.LTLEnv(env, ltl_sampler="Default")
agent = RandomAgent(env.action_space)
elif (args["command"] == "viz"):
# If the config is available (from trainig) then just load it here instead of asking the user of this script to provide all training time configs
config = vars(utils.load_config(args["model_path"]))
args.update(config)