MAX_COLOR = 25.
# Set up paths for logging the data from the simulation run
if not os.path.exists('./figures/' + str(reward_function)):
os.makedirs('./figures/' + str(reward_function))
logging.basicConfig(filename = './figures/'+ reward_function + '/robot.log', level = logging.INFO)
logger = logging.getLogger('robot')
# Create a random enviroment sampled from a GP with an RBF kernel and specified hyperparameters, mean function 0
# The enviorment will be constrained by a set of uniformly distributed sample points of size NUM_PTS x NUM_PTS
ranges = (0., 10., 0., 10.)
world = envlib.Environment(ranges = ranges,
NUM_PTS = 20,
variance = 100.0,
lengthscale = 1.0,
visualize = True,
seed = seed,
MIN_COLOR=MIN_COLOR,
MAX_COLOR=MAX_COLOR)
# Create the evaluation class used to quantify the simulation metrics
evaluation = evalib.Evaluation(world = world, reward_function = reward_function)
# Populate a world with obstacles
# ow = obslib.FreeWorld()
# ow = obslib.BlockWorld(ranges, 1, dim_blocks= (2,2), centers=[(7,7)])
ow = obslib.ChannelWorld(ranges, (2.5,7), 3, 0.2)
# Create the point robot
robot = roblib.Robot(sample_world = world.sample_value, #function handle for collecting observations
start_loc = (5.0, 5.0, 0.0), #where robot is instantiated
logging.basicConfig(filename = './figures/'+ REWARD_FUNCTION + '/robot.log', level = logging.INFO)
logger = logging.getLogger('robot')
#设置world的范围
ranges = (0., 10., 0., 10.)
#加载无障碍物环境
ow = obslib.FreeWorld() #a world without obstacles
#创建机器人能够识别的环境
#没有观测值的预设先验的环境更新
world = envlib.Environment(ranges = ranges,
NUM_PTS = 20,
variance = 100.0,
lengthscale = 1.0,
visualize = False,
seed = SEED,
MIN_COLOR=MIN_COLOR,
MAX_COLOR=MAX_COLOR,
obstacle_world = ow,
noise=10.0)
#将更新好后的地图和奖励函数类型
# Create the evaluation class used to quantify the simulation metrics
#只初始化参数没有动作
evaluation = evalib.Evaluation(world = world, reward_function = REWARD_FUNCTION)
#设置机器人参数
robot = roblib.Robot(sample_world = world.sample_value, #function handle for collecting observations
start_loc = (5.0, 5.0, 0.0), #where robot is instantiated
dimension = 2,
extent = ranges, #extent of the explorable environment
kernel_file = None,
NOISE = 0.19836 / 10.0
else:
gp_world = None
# VAR = 50.0
# LEN = 5.0
# NOISE = 0.1
VAR = 100.0
LEN = 1.0
NOISE = 1.0
world = envlib.Environment(ranges = ranges,
NUM_PTS = 100,
variance = VAR,
lengthscale = LEN,
noise = NOISE,
visualize = True,
seed = SEED,
MAX_COLOR = MAX_COLOR,
MIN_COLOR = MIN_COLOR,
model = gp_world,
obstacle_world = ow)
# Create the evaluation class used to quantify the simulation metrics
evaluation = evalib.Evaluation(world = world, reward_function = REWARD_FUNCTION)
# Generate a prior dataset
'''
x1observe = np.linspace(ranges[0], ranges[1], 5)
x2observe = np.linspace(ranges[2], ranges[3], 5)
x1observe, x2observe = np.meshgrid(x1observe, x2observe, sparse = False, indexing = 'xy')
data = np.vstack([x1observe.ravel(), x2observe.ravel()]).T
# The enviorment will be constrained by a set of uniformly distributed sample points of size NUM_PTS x NUM_PTS
ranges = (0., 10., 0., 10.)
# Create obstacle world
ow = obslib.FreeWorld()
# ow = obslib.ChannelWorld(ranges, (3.5, 7.), 3., 0.3)
# ow = obslib.BugTrap(ranges, (2.2, 3.0), 4.6, orientation = 'left', width = 5.0)
# ow = obslib.BlockWorld(ranges,12, dim_blocks=(1., 1.), centers=[(2.5, 2.5), (7.,4.), (5., 8.), (8.75, 6.), (3.5,6.), (6.,1.5), (1.75,5.), (6.2,6.), (8.,8.5), (4.2, 3.8), (8.75,2.5), (2.2,8.2)])
world = envlib.Environment(
ranges=ranges,
NUM_PTS=20,
variance=
100.0, # TODO: why is only the lengthscale a vector for asymmetric kernels?
lengthscale=LENGTHSCALE,
noise=0.5,
dim=DIM,
visualize=True,
seed=SEED,
MIN_COLOR=MIN_COLOR,
MAX_COLOR=MAX_COLOR,
obstacle_world=ow,
time_duration=DURATION)
# noise= 5.0)
# Create the evaluation class used to quantify the simulation metrics
evaluation = evalib.Evaluation(world=world, reward_function=REWARD_FUNCTION)
# Generate a prior dataset
x1observe = np.linspace(ranges[0], ranges[1], 20)
x2observe = np.linspace(ranges[2], ranges[3], 20)
x1observe, x2observe = np.meshgrid(x1observe,