def main():
core_dir = "./environment"
envs_dir = './environment/envs',
xmls_dir = './environment/assets/xmls',
examine_env("./environment/envs/icra.py", {},
core_dir=core_dir, envs_dir=envs_dir, xmls_dir=xmls_dir,
env_viewer=EnvViewer)
def main():
core_dir = "./"
envs_dir = './envs',
xmls_dir = './assets/xmls',
examine_env("./envs/base.py", {},
core_dir=core_dir,
envs_dir=envs_dir,
xmls_dir=xmls_dir,
env_viewer=EnvViewer)
def main(argv):
'''
examine.py is used to display environments
Example uses:
bin/examine.py simple_particle
bin/examine.py examples/particle_gather.py
bin/examine.py particle_gather n_food=5 floorsize=5
bin/examine.py example_env_examine.jsonnet
'''
env_names, env_kwargs = parse_arguments(argv)
assert len(
env_names) == 1, 'You must provide exactly 1 environment to examine.'
env_name = env_names[0]
examine_env(env_name,
env_kwargs,
core_dir=worldgen_path(),
envs_dir='examples',
xmls_dir='xmls',
env_viewer=EnvViewer)
print(main.__doc__)
def main(argv):
'''
examine.py is used to display environments and run policies.
For an example environment jsonnet, see
mujoco-worldgen/examples/example_env_examine.jsonnet
You can find saved policies and the in the 'examples' together with the environment they were
trained in and the hyperparameters used. The naming used is 'examples/<env_name>.jsonnet' for
the environment jsonnet file and 'examples/<env_name>.npz' for the policy weights file.
Example uses:
bin/examine.py hide_and_seek
bin/examine.py mae_envs/envs/base.py
bin/examine.py base n_boxes=6 n_ramps=2 n_agents=3
bin/examine.py my_env_jsonnet.jsonnet
bin/examine.py my_env_jsonnet.jsonnet my_policy.npz
bin/examine.py hide_and_seek my_policy.npz n_hiders=3 n_seekers=2 n_boxes=8 n_ramps=1
'''
names, kwargs = parse_arguments(argv)
env_name = names[0]
core_dir = abspath(join(dirname(__file__), '..'))
envs_dir = 'mae_envs/envs',
xmls_dir = 'xmls',
if len(names) == 1: # examine the environment
examine_env(env_name,
kwargs,
core_dir=core_dir,
envs_dir=envs_dir,
xmls_dir=xmls_dir,
env_viewer=EnvViewer)
if len(names) >= 2: # run policies on the environment
# importing PolicyViewer and load_policy here because they depend on several
# packages which are only needed for playing policies, not for any of the
# environments code.
from mae_envs.viewer.policy_viewer import PolicyViewer
from ma_policy.load_policy import load_policy
policy_names = names[1:]
env, args_remaining_env = load_env(env_name,
core_dir=core_dir,
envs_dir=envs_dir,
xmls_dir=xmls_dir,
return_args_remaining=True,
**kwargs)
if isinstance(env.action_space, Tuple):
env = JoinMultiAgentActions(env)
if env is None:
raise Exception(
f'Could not find environment based on pattern {env_name}')
env.reset() # generate action and observation spaces
assert np.all([name.endswith('.npz') for name in policy_names])
policies = [
load_policy(name, env=env, scope=f'policy_{i}')
for i, name in enumerate(policy_names)
]
args_remaining_policy = args_remaining_env
if env is not None and policies is not None:
args_to_pass, args_remaining_viewer = extract_matching_arguments(
PolicyViewer, kwargs)
args_remaining = set(args_remaining_env)
args_remaining = args_remaining.intersection(
set(args_remaining_policy))
args_remaining = args_remaining.intersection(
set(args_remaining_viewer))
assert len(args_remaining) == 0, (
f"There left unused arguments: {args_remaining}. There shouldn't be any."
)
viewer = PolicyViewer(env, policies, **args_to_pass)
viewer.run()
print(main.__doc__)
def main(args):
'''
examine.py is used to display environments and run policies.
For an example environment jsonnet, see
mujoco-worldgen/examples/example_env_examine.jsonnet
You can find saved policies and the in the 'examples' together with the environment they were
trained in and the hyperparameters used. The naming used is 'examples/<env_name>.jsonnet' for
the environment jsonnet file and 'examples/<env_name>.npz' for the policy weights file.
Example uses:
bin/examine.py hide_and_seek
bin/examine.py mae_envs/envs/base.py
bin/examine.py base n_boxes=6 n_ramps=2 n_agents=3
bin/examine.py my_env_jsonnet.jsonnet
bin/examine.py my_env_jsonnet.jsonnet my_policy.npz
bin/examine.py hide_and_seek my_policy.npz n_hiders=3 n_seekers=2 n_boxes=8 n_ramps=1
bin/examine.py examples/hide_and_seek_quadrant.jsonnet examples/hide_and_seek_quadrant.npz
'''
#names, kwargs = parse_arguments(argv)
parser = get_config()
args = parse_args(args, parser)
kwargs={'args': args}
env_name = args.env_name
num_hiders = args.num_hiders
num_seekers = args.num_seekers
num_agents = num_hiders + num_seekers
core_dir = abspath(join(dirname(__file__)))
envs_dir = 'envs/hns/envs' # where hide_and_seek.py is.
xmls_dir = 'xmls'
if args.use_render: # run policies on the environment
# importing PolicyViewer and load_policy here because they depend on several
# packages which are only needed for playing policies, not for any of the
# environments code.
from onpolicy.envs.hns.viewer.policy_viewer import PolicyViewer_hs
from onpolicy.envs.hns.ma_policy.load_policy import load_policy
env, args_remaining_env = load_env(env_name, core_dir=core_dir,
envs_dir=envs_dir, xmls_dir=xmls_dir,
return_args_remaining=True, **kwargs)
if isinstance(env.action_space, Tuple):
env = JoinMultiAgentActions(env)
if env is None:
raise Exception(f'Could not find environment based on pattern {env_name}')
env.reset() # generate action and observation spaces
policies = []
for agent_id in range(num_agents):
if args.share_policy:
actor_critic = torch.load(str(args.model_dir) + "/agent_model.pt")['model']
else:
actor_critic = torch.load(str(args.model_dir) + "/agent" + str(agent_id) + "_model.pt")['model']
policies.append(actor_critic)
args_remaining_policy = args_remaining_env
if env is not None and policies is not None:
args_to_pass, args_remaining_viewer = extract_matching_arguments(PolicyViewer_hs, kwargs)
args_remaining = set(args_remaining_env)
args_remaining = args_remaining.intersection(set(args_remaining_policy))
args_remaining = args_remaining.intersection(set(args_remaining_viewer))
assert len(args_remaining) == 0, (
f"There left unused arguments: {args_remaining}. There shouldn't be any.")
viewer = PolicyViewer_hs(env, policies, **args_to_pass)
viewer.run()
else:
# examine the environment
examine_env(env_name, kwargs,
core_dir=core_dir, envs_dir=envs_dir, xmls_dir=xmls_dir,
env_viewer=EnvViewer)
