def _partial_regen(self, n_new_sequences=1):
if self.regen == "navreptrain":
env = NavRepTrainEnv(silent=True, scenario='train', adaptive=False)
env.soadrl_sim.human_num = 20
data = generate_vae_dataset(
env,
n_sequences=n_new_sequences,
policy=ORCAPolicy(suicide_if_stuck=True),
render=False,
archive_dir=None)
if self.pre_convert_obs:
data["obs"] = scans_to_lidar_obs(data["scans"],
self.lidar_mode,
self.rings_def,
self.channel_first)
else:
print("Regen {} failed".format(self.regen))
return
for k in self.data.keys():
N = len(data[k]) # should be the same for each key
# check end inside loop to avoid having to pick an arbitrary key
if self.regen_head_index + N > len(self.data[k]):
self.regen_head_index = 0
# replace data
i = self.regen_head_index
self.data[k][i:i + N] = data[k]
self.regen_head_index += N
from navrep.envs.navreptrainenv import NavRepTrainEnv
from navrep.tools.commonargs import parse_common_args
from navrep.scripts.test_navrep import run_test_episodes
class LuciaPolicy(object):
""" legacy SOADRL policy from lucia's paper, takes in agents state, local map """
def __init__(self, env):
self.env = env
def act(self, obs):
state, local_map = obs
return self.env.soadrl_sim.robot.act(state, local_map)
if __name__ == '__main__':
args, _ = parse_common_args()
env = NavRepTrainEnv(silent=True, scenario='test', legacy_mode=True)
policy = LuciaPolicy(env)
run_test_episodes(env, policy, render=args.render)
import os
from navrep.tools.envplayer import EnvPlayer
from navrep.envs.navreptrainenv import NavRepTrainEnv
if __name__ == "__main__":
os.environ["CUDA_VISIBLE_DEVICES"] = "-1" # disable GPU
env = NavRepTrainEnv()
player = EnvPlayer(env)
import numpy as np
from timeit import default_timer as timer
from tqdm import tqdm
from navrep.envs.navreptrainenv import NavRepTrainEnv
from navrep.tools.commonargs import parse_common_args
if __name__ == "__main__":
args, _ = parse_common_args()
n = args.n
if n is None:
n = 1000000
env = NavRepTrainEnv(scenario='train',
silent=True,
adaptive=False,
collect_statistics=False)
env.reset()
action = np.array([0., 0., 0.])
tic = timer()
for i in tqdm(range(n)):
env.step(action)
toc = timer()
elapsed = toc - tic
print("Executed {} simulation steps in {:.1f} seconds.".format(n, elapsed))
archive_dir = os.path.expanduser("~/navrep/datasets/V/marktwo")
if args.dry_run:
archive_dir = "/tmp/navrep/datasets/V/marktwo"
env = MarkEnv(silent=True, maps=SECOND_TRAIN_MAPS)
generate_vae_dataset(env,
n_sequences=n_sequences,
subset_index=args.subproc_id,
n_subsets=args.n_subprocs,
render=args.render,
archive_dir=archive_dir)
if args.environment == "navreptrain":
archive_dir = os.path.expanduser("~/navrep/datasets/V/navreptrain")
if args.dry_run:
archive_dir = "/tmp/navrep/datasets/V/navreptrain"
env = NavRepTrainEnv(silent=True,
scenario='train',
adaptive=False,
collect_statistics=False)
env.soadrl_sim.human_num = 20
generate_vae_dataset(env,
n_sequences=n_sequences,
subset_index=args.subproc_id,
n_subsets=args.n_subprocs,
policy=ORCAPolicy(suicide_if_stuck=True),
render=args.render,
archive_dir=archive_dir)
if args.environment == "irl":
folder_to_archive(
directory="~/rosbags/iros_rosbags",
archive_dir=os.path.expanduser("~/navrep/datasets/V/irl"),
)