def find_best_k_views(self, eye_pos, all_dist, all_pos, avoid_block=False):
least_order = (np.argsort(all_dist))
top_k = []
num_to_test = self.k * 2
curr_num = 0
all_pos = np.array(all_pos)
if not avoid_block:
return least_order[:self.k]
while len(top_k) < self.k:
curr_order = least_order[curr_num: curr_num + num_to_test]
curr_pos = all_pos[curr_order]
print("Curr num", curr_num, "top k", len(top_k), self.k)
if len(curr_pos) <= p.MAX_RAY_INTERSECTION_BATCH_SIZE:
collisions = list(p.rayTestBatch([eye_pos] * len(curr_pos), curr_pos))
else:
collisions = []
curr_i = 0
while (curr_i < len(curr_pos)):
curr_n = min(len(curr_pos), curr_i + p.MAX_RAY_INTERSECTION_BATCH_SIZE - 1)
collisions = collisions + list(p.rayTestBatch([eye_pos] * (curr_n - curr_i), curr_pos[curr_i: curr_n]))
curr_i = curr_n
has_collision = [c[0] > 0 for c in collisions]
## (hzyjerry): ray_casting-based view selection occasionally gives unstable behaviour. Will keep watching on this
for i, x in enumerate(curr_order):
if not has_collision[i]:
top_k.append(x)
return top_k
def find_best_k_views(self, eye_pos, all_dist, all_pos):
least_order = (np.argsort(all_dist))
if len(all_pos) <= p.MAX_RAY_INTERSECTION_BATCH_SIZE:
collisions = list(p.rayTestBatch([eye_pos] * len(all_pos), all_pos))
else:
collisions = []
curr_i = 0
while (curr_i < len(all_pos)):
curr_n = min(len(all_pos), curr_i + p.MAX_RAY_INTERSECTION_BATCH_SIZE - 1)
collisions = collisions + list(p.rayTestBatch([eye_pos] * (curr_n - curr_i), all_pos[curr_i: curr_n]))
curr_i = curr_n
collisions = [c[0] for c in collisions]
top_k = []
for i in range(len(least_order)):
if len(top_k) >= self.k:
break
## (hzyjerry): disabling ray_casting-based view selection because it gives unstable behaviour right now.
top_k.append(least_order[i])
if len(top_k) < self.k:
for o in least_order:
if o not in top_k:
top_k.append(o)
if len(top_k) >= self.k:
break
return top_k
def get_obs(self, env):
"""
Get current LiDAR sensor reading and occupancy grid (optional)
:return: LiDAR sensor reading and local occupancy grid, normalized to [0.0, 1.0]
"""
laser_angular_half_range = self.laser_angular_range / 2.0
if self.laser_link_name not in env.robots[0].parts:
raise Exception('Trying to simulate LiDAR sensor, but laser_link_name cannot be found in the robot URDF file. Please add a link named laser_link_name at the intended laser pose. Feel free to check out assets/models/turtlebot/turtlebot.urdf and examples/configs/turtlebot_p2p_nav.yaml for examples.')
laser_pose = env.robots[0].parts[self.laser_link_name].get_pose()
angle = np.arange(
-laser_angular_half_range / 180 * np.pi,
laser_angular_half_range / 180 * np.pi,
self.laser_angular_range / 180.0 * np.pi / self.n_horizontal_rays)
unit_vector_local = np.array(
[[np.cos(ang), np.sin(ang), 0.0] for ang in angle])
transform_matrix = quat2mat(
[laser_pose[6], laser_pose[3], laser_pose[4], laser_pose[5]]) # [x, y, z, w]
unit_vector_world = transform_matrix.dot(unit_vector_local.T).T
start_pose = np.tile(laser_pose[:3], (self.n_horizontal_rays, 1))
start_pose += unit_vector_world * self.min_laser_dist
end_pose = laser_pose[:3] + unit_vector_world * self.laser_linear_range
results = p.rayTestBatch(start_pose, end_pose, 6) # numThreads = 6
# hit fraction = [0.0, 1.0] of self.laser_linear_range
hit_fraction = np.array([item[2] for item in results])
hit_fraction = self.noise_model.add_noise(hit_fraction)
scan = np.expand_dims(hit_fraction, 1)
state = {}
state['scan'] = scan
if 'occupancy_grid' in self.modalities:
state['occupancy_grid'] = self.get_local_occupancy_grid(scan)
return state
def pybullet_path_checking(start, end):
'''
:param start: starting coordinate in real world
:return: True:no collision with object in path
False: collide with object in path
'''
pos = end
dz = 2
dense_of_ray = 1 #describ how dense the ray for collision detection
_, _, _, hit_position, _ = p.rayTest(
start, [start[0], start[1], start[2] - dz])[0]
height = start[2] - hit_position[2]
start_pos_array = [[
start[0], start[1], hit_position[2] + i * height / dense_of_ray
] for i in range(1, dense_of_ray + 1)]
_, _, _, end_hit_position, _ = p.rayTest(pos,
[pos[0], pos[1], pos[2] - dz])[0]
end_height = pos[2] - end_hit_position[2]
end_pos_array = [[
pos[0], pos[1], end_hit_position[2] + i * end_height / dense_of_ray
] for i in range(1, dense_of_ray + 1)]
result = p.rayTestBatch(start_pos_array, end_pos_array)
for r in result:
if r[1] == -1:
#no collision
continue
else:
return False
return True
def rayTest(robot_id: int,
ray_length: float,
ray_num: int = 5,
physicsClientId: int = 0):
"""
:param robot_id: 不多说
:param ray_length: 激光长度
:param ray_num: 激光数量(需要说明，激光头均匀分布在-pi/2到pi/2之间)
"""
basePos, baseOrientation = p.getBasePositionAndOrientation(
robot_id, physicsClientId=physicsClientId)
matrix = p.getMatrixFromQuaternion(baseOrientation,
physicsClientId=physicsClientId)
basePos = np.array(basePos)
matrix = np.array(matrix).reshape([3, 3])
# 选定在机器人的本地坐标系中中心到几个激光发射点的向量
# 此处的逻辑为先计算出local坐标系中的距离单位向量，再变换到世界坐标系中
unitRayVecs = np.array([[cos(alpha), sin(alpha), 0]
for alpha in np.linspace(-np.pi / 2., np.pi /
2., ray_num)])
unitRayVecs = unitRayVecs.dot(matrix.T)
# 通过广播运算得到世界坐标系中所有激光发射点的坐标
rayBegins = basePos + BASE_RADIUS * unitRayVecs
rayTos = rayBegins + ray_length * unitRayVecs
results = p.rayTestBatch(rayBegins,
rayTos,
physicsClientId=physicsClientId)
return rayBegins, rayTos, results
def get_scan(self):
"""
:return: LiDAR sensor reading, normalized to [0.0, 1.0]
"""
laser_angular_half_range = self.laser_angular_range / 2.0
if self.laser_link_name not in self.robots[0].parts:
raise Exception('Trying to simulate LiDAR sensor, but laser_link_name cannot be found in the robot URDF file. Please add a link named laser_link_name at the intended laser pose. Feel free to check out assets/models/turtlebot/turtlebot.urdf and examples/configs/turtlebot_p2p_nav.yaml for examples.')
laser_pose = self.robots[0].parts[self.laser_link_name].get_pose()
angle = np.arange(-laser_angular_half_range / 180 * np.pi,
laser_angular_half_range / 180 * np.pi,
self.laser_angular_range / 180.0 * np.pi / self.n_horizontal_rays)
unit_vector_local = np.array([[np.cos(ang), np.sin(ang), 0.0] for ang in angle])
transform_matrix = quat2mat([laser_pose[6], laser_pose[3], laser_pose[4], laser_pose[5]]) # [x, y, z, w]
unit_vector_world = transform_matrix.dot(unit_vector_local.T).T
start_pose = np.tile(laser_pose[:3], (self.n_horizontal_rays, 1))
start_pose += unit_vector_world * self.min_laser_dist
end_pose = laser_pose[:3] + unit_vector_world * self.laser_linear_range
results = p.rayTestBatch(start_pose, end_pose, 6) # numThreads = 6
hit_fraction = np.array([item[2] for item in results]) # hit fraction = [0.0, 1.0] of self.laser_linear_range
hit_fraction = self.add_naive_noise_to_sensor(hit_fraction, self.scan_noise_rate)
scan = np.expand_dims(hit_fraction, 1)
xyz = hit_fraction[:, np.newaxis] * unit_vector_local * 10
xyz = xyz[np.equal(np.isnan(xyz), False)]
xyz = xyz.reshape(xyz.shape[0] // 3, -1)
return xyz#scan
def _laserScan(self):
"""
INTERNAL METHOD, a loop that simulate the laser and update the distance
value of each laser
"""
lastLidarTime = time.time()
while not self._module_termination:
nowLidarTime = time.time()
if (nowLidarTime - lastLidarTime > 1 / LASER_FRAMERATE):
results = pybullet.rayTestBatch(
self.ray_from,
self.ray_to,
parentObjectUniqueId=self.robot_model,
parentLinkIndex=self.laser_id,
physicsClientId=self.physics_client)
for i in range(NUM_RAY * len(ANGLE_LIST_POSITION)):
hitObjectUid = results[i][0]
hitFraction = results[i][2]
hitPosition = results[i][3]
self.laser_value[i] = hitFraction * RAY_LENGTH
if self.display:
if not self.ray_ids:
self._createDebugLine()
if (hitFraction == 1.):
pybullet.addUserDebugLine(
self.ray_from[i],
self.ray_to[i],
RAY_MISS_COLOR,
replaceItemUniqueId=self.ray_ids[i],
parentObjectUniqueId=self.robot_model,
parentLinkIndex=self.laser_id,
physicsClientId=self.physics_client)
else: # pragma: no cover
localHitTo = [
self.ray_from[i][0] + hitFraction *
(self.ray_to[i][0] - self.ray_from[i][0]),
self.ray_from[i][1] + hitFraction *
(self.ray_to[i][1] - self.ray_from[i][1]),
self.ray_from[i][2] + hitFraction *
(self.ray_to[i][2] - self.ray_from[i][2])
]
pybullet.addUserDebugLine(
self.ray_from[i],
localHitTo,
RAY_HIT_COLOR,
replaceItemUniqueId=self.ray_ids[i],
parentObjectUniqueId=self.robot_model,
parentLinkIndex=self.laser_id,
physicsClientId=self.physics_client)
else:
if self.ray_ids:
self._resetDebugLine()
lastLidarTime = nowLidarTime
def single_test(urdf_file, isDIRECT=True):
p.connect(p.DIRECT if isDIRECT else p.GUI)
p.setAdditionalSearchPath(pybullet_data.getDataPath())
p.loadURDF(urdf_file, [3, 3, 1])
numRays = 1024
rayLen = 13
info_lst = []
rayFrom_lst = []
rayTo_lst = []
for i in range(numRays):
rayFrom = [0, 0, 1]
rayTo = [
rayLen * math.sin(2. * math.pi * float(i) / numRays),
rayLen * math.cos(2. * math.pi * float(i) / numRays), 1
]
info = p.rayTest(rayFrom, rayTo)
info_lst.append(info[0])
p.addUserDebugLine(rayFrom, rayTo, [1, 0, 0])
rayFrom_lst.append(rayFrom)
rayTo_lst.append(rayTo)
info_lst_batch = p.rayTestBatch(rayFrom_lst, rayTo_lst)
collide_predicate = lambda info: info[0] != -1
info_filtered = list(filter(collide_predicate, info_lst))
info_batch_filtered = list(filter(collide_predicate, info_lst_batch))
p.disconnect()
return info_filtered, info_batch_filtered
def compute_points_from_camera(self, camera, sample_size, max_range, sigma,
sigma_pure, dt):
#p.removeAllUserDebugItems()
ray_from = []
ray_to = []
rand_x = np.random.normal(camera.center[0], sigma, sample_size)
rand_x = np.insert(
rand_x, 0,
camera.center[0]) # insert center for center of attention
rand_y = np.random.normal(camera.center[1], sigma, sample_size)
rand_y = np.insert(
rand_y, 0,
camera.center[1]) # insert center for center of attention
target_matrix = np.zeros((4, sample_size + 1))
target_matrix[0, :] = -0.5
target_matrix[1, :] = 0.5
target_matrix[2, :] = rand_x
target_matrix[3, :] = -rand_y
ortho = camera.trans_mat.dot(target_matrix)
rayTo = (ortho +
(camera.forward - camera.position)[:, np.newaxis]) * max_range
for s in range(sample_size):
target = rayTo[:, s]
ray_from.append(camera.position)
ray_to.append(target)
hits = p.rayTestBatch(ray_from, ray_to)
point_clouds = {}
element_center_attention = hits[0][0] # get object id
center_point = hits[0][3] # get hit position
max_radius = 1.0
# name, pos, intensity
selected_hits = self.compute_intensity(hits[0],
hits[1:],
max_radius,
gamma=0.5,
penalty=1.0,
dt=dt)
for h in selected_hits:
#p.addUserDebugLine(camera.position, [h[1], h[2], h[3]], rayMissColor)
if h[0] not in point_clouds.keys():
point_clouds[h[0]] = []
point_clouds[h[0]].append(h[1:])
return point_clouds, element_center_attention, center_point
def test_smallsphere(pos_):
pos = np.array(pos_)
ss = p.loadURDF("r2d2.urdf")
p.resetBasePositionAndOrientation(ss, pos, [0, 0, 0, 1])
margin = np.array([0.5, 0.0, 0.0])
p_start = pos - margin
p_end = pos + margin
p.addUserDebugLine(p_start, p_end, [1, 0, 0])
input("press")
result = p.rayTestBatch([p_start], [p_end])
return result
def read_laser_scan(self): numRays=10 # rayFrom,rayTo = self.init_laserscanner() numThreads=0 results = p.rayTestBatch(self.rayFrom,self.rayTo,numThreads, parentObjectUniqueId=self.robot_uid, parentLinkIndex=self.hokuyo_joint) result = [] trans,rot = p.getBasePositionAndOrientation(self.robot_uid) for i in range (numRays): laser = np.linalg.norm(np.array([trans[0],trans[1]]) - np.array([results[i][3][0],results[i][3][1]])) result.append(laser) return np.array([result])
def batch_ray_collision(rays, threads=1):
assert 1 <= threads <= p.MAX_RAY_INTERSECTION_BATCH_SIZE
if not rays:
return []
step_simulation() # Needed for some reason
ray_starts = [start for start, _ in rays]
ray_ends = [end for _, end in rays]
return [RayResult(*tup) for tup in p.rayTestBatch(
ray_starts, ray_ends,
numThreads=threads,
#parentObjectUniqueId=
#parentLinkIndex=
physicsClientId=CLIENT)]
def compute(self, world, state):
import pybullet as pb
object_translation_rotation = []
# camera.py ImageData namedtuple
camera = world.cameras[0]
image_data = camera.capture()
image_data_arrays = [np.array(value) for value in image_data] + \
[camera.matrix, camera.projection_matrix]
# 'camera_view_matrix' namedtuple index is 4
# TODO(ahundt) ensure camera matrix translation component is from world origin to camera origin
# camera ray is from the origin of the camera
camera_transform_array = np.transpose(image_data[4]).reshape((4, 4))
# use the more conveniently formatted transform for writing out
image_data_arrays[4] = camera_transform_array
camera_translation = camera_transform_array[0:3, 3].tolist()
# print("TEST IMAGEDATA named tuple img matrix: \n", )
# print("TEST IMAGEDATA named tuple img matrix translation: ", camera_translation)
camera_ray_from = [camera_translation] * len(
world.id_by_object.items())
# camera_ray_to is the center of each object
camera_ray_to = []
for name, oid in world.id_by_object.items():
# print("oid type:", str(type(oid)))
# print("actor type:", str(type(world.actors[oid])))
obj = world.actors[oid].getState()
p = obj.T.p
q = obj.T.M.GetQuaternion()
one_t_r = np.array([p[0], p[1], p[2], q[0], q[1], q[2], q[3]],
dtype=np.float32)
object_translation_rotation.append(one_t_r)
camera_ray_to.append(list(obj.T.p))
# print("lengths: ", len(camera_ray_from), len(camera_ray_to))
object_surface_points = []
# TODO(ahundt) allow multiple simulations to run
raylist = pb.rayTestBatch(camera_ray_from, camera_ray_to)
for i, (uid, linkidx, hitfrac, hitpos,
hitnormal) in enumerate(raylist):
if uid is -1:
# if the object wasn't hit, use its origin
name, oid = world.id_by_object.items()[i]
obj = world.actors[oid].getState()
object_surface_points += [obj.T.p]
else:
object_surface_points += hitpos
return [
np.array(object_translation_rotation),
np.array(state.arm),
np.array(state.gripper)
] + image_data_arrays[1:] + [np.array(object_surface_points)]
def draw_rays(self):
'''
draws rays to be visualized in sim
'''
ray_to = np.zeros([self.ray_count, 3])
ray_ids = np.zeros(self.ray_count)
ray_angles = np.mod(self.yaw + self.rays, 2 * np.pi)
for ii, th in enumerate(ray_angles):
ray_to[ii] = self.x+ np.array([self.ray_len*np.cos(th),\
self.ray_len*np.sin(th),0])
if self.debug:
p.addUserDebugLine(self.x, ray_to[ii], self.ray_miss_color)
return p.rayTestBatch(self.ray_from, ray_to)
def getlocalMapObservation(self):
### raycast around the area of the agent
"""
For now this includes the agent in the center of the computation
"""
import numpy as np
pos = p.getBasePositionAndOrientation(self._agent)[0]
### number of samples
size = 16
### width of box
dimensions = 8.0
toRays = []
for i in range(0, size):
for j in range(0, size):
toRays.append([(1.0 / (size * 1.0)) * i * dimensions,
(1.0 / (size * 1.0)) * j * dimensions, 0])
assert (len(toRays) == (size * size))
toRays = np.array(toRays)
### Adjust to put agent in middle of map
toRays = toRays + pos - np.array(
[dimensions / 2.0, dimensions / 2.0, 0])
# print ("toRays:", toRays )
fromRays = toRays + np.array([0, 0, 5])
rayResults = p.rayTestBatch(fromRays, toRays)
intersections = [ray[0] for ray in rayResults]
# print (intersections)
### fix intersections that could be goal
for ray in range(len(intersections)):
if (intersections[ray] in [self._target, self._agent]):
# print ("bad index: ", ray)
intersections[ray] = -1
# bad_indecies = np.where(intersections == self._target)[0]
# print ("bad_indecies: ", bad_indecies)
# bad_indecies = np.where(intersections == int(self._agent))
# print ("bad_indecies: ", bad_indecies)
# print ("self._agent: ", self._agent)
"""
if ( len(bad_indecies) > 0):
# print ("self._target: ", self._target)
intersections[bad_indecies] = -1
"""
# intersections_ = np.reshape(intersections, (size, size))
# print ("intersections", intersections_)
intersections = np.array(np.greater(intersections, 0), dtype="int")
return intersections
def get_range_reading(theta):
ranges = []
rayFroms = [lidar_pos for i in range(num_rays)]
rayTos = get_batch_ray_to(theta)
rayTestInfo = p.rayTestBatch(rayFroms, rayTos)
for i in range(len(rayTestInfo)):
# compute hit distance
d = float(rayTestInfo[i][2]) * lidar_range
# if no hit
if rayTestInfo[i][0] == -1:
d = lidar_range
ranges.append(1 - d / lidar_range)
# Draw debug rays
for i in range(0, len(rayTos), 5):
p.addUserDebugLine(lidar_pos, rayTos[i], [1, 0, 0], lifeTime=0.6)
return ranges
def get_lidar_scan(self):
if self.valid_buffer_scan:
return self.last_scan
'''
Obtain lidar scan values for current state.
Returns:
list: Scan values for range and resolution specified in
params dict.
'''
scan_front = None
scan_rear = None
# Get pose of lidar
states = pb.getLinkState(self.robotId, self.lidarLinkId1, False, False,
self.clientId)
lidar_pos, lidar_ori = states[4:6]
lidar_pos = np.array(lidar_pos)
R = np.array(pb.getMatrixFromQuaternion(lidar_ori))
R = np.reshape(R, (3, 3))
scan_l1 = R.dot(self.rays[0]).T + lidar_pos
scan_h1 = R.dot(self.rays[1]).T + lidar_pos
# Get pose of lidar
states = pb.getLinkState(self.robotId, self.lidarLinkId2, False, False,
self.clientId)
lidar_pos, lidar_ori = states[4:6]
lidar_pos = np.array(lidar_pos)
R = np.array(pb.getMatrixFromQuaternion(lidar_ori))
R = np.reshape(R, (3, 3))
scan_l2 = R.dot(self.rays[0]).T + lidar_pos
scan_h2 = R.dot(self.rays[1]).T + lidar_pos
scan_l = np.concatenate((scan_l1, scan_l2))
scan_h = np.concatenate((scan_h1, scan_h2))
scan_r = pb.rayTestBatch(scan_l.tolist(), scan_h.tolist(),
self.clientId)
scan = [x[2] * self.p["sensors"]["lidar"]["range"] for x in scan_r]
scan_front = scan[:len(scan_l1)]
scan_rear = scan[len(scan_l1):]
self.last_scan = [scan_front, scan_rear]
self.valid_buffer_scan = True
return [scan_front, scan_rear]
def cast_rays_spherical(self, origin, azimuth_from, azimuth_to,
azimuth_res, polar_from, polar_to, polar_res,
max_ray_len):
def interp(frm, to, i, res):
return frm + (to - frm) * (i / float(max(res - 1, 1)))
ray_count = polar_res * azimuth_res
to_pts = [None] * ray_count
ray_i = 0
for pol_i in range(polar_res):
el = interp(polar_from, polar_to, pol_i, polar_res)
rcos_theta = max_ray_len * np.cos(el)
z = max_ray_len * np.sin(el)
for azi_i in range(azimuth_res):
az = interp(azimuth_from, azimuth_to, azi_i, azimuth_res)
x = rcos_theta * np.cos(az)
y = rcos_theta * np.sin(az)
to_pts[ray_i] = (x, y, z)
ray_i = ray_i + 1
results = []
tested_ray_count = 0
while ray_count - tested_ray_count > 0:
# the '-1' is due to seemingly a bug in pybullet, which seems to return
# one result less than the batch size, when using the full batch size
batch_size = min(ray_count - tested_ray_count,
bullet.MAX_RAY_INTERSECTION_BATCH_SIZE - 1)
batch_results = bullet.rayTestBatch(
[origin] * batch_size,
to_pts[tested_ray_count:tested_ray_count + batch_size])
results.extend(batch_results)
tested_ray_count = tested_ray_count + batch_size
hits = [None] * ray_count
hit_i = 0
for i in range(ray_count):
hit_uid = results[i][0]
if hit_uid >= 0:
hit_pos = results[i][3]
hits[hit_i] = (hit_uid, hit_pos)
hit_i = hit_i + 1
return hits[:hit_i]
def observe(self) -> NDArray[(Any, ), np.float]:
results = p.rayTestBatch(self._from,
self._to,
0,
parentObjectUniqueId=self.body_id,
parentLinkIndex=self.joint_index)
hit_fractions = np.array(results, dtype=object)[:,
2].astype(dtype=float)
ranges = self._config.range * hit_fractions + self._config.min_range
noise = np.random.uniform(1.0 - self._config.accuracy,
1.0 + self._config.accuracy,
size=ranges.shape)
scan = np.clip(ranges * noise,
a_min=self._config.min_range,
a_max=self._config.range)
if self._config.debug:
self._display_rays(hit_fractions, scan)
return scan
def detectLaserCollisions(self):
'''Detect collision of laser rays and render it on GUI'''
for body in self.bodies:
if body.hasLaser:
for i in range(len(body.lasers)):
(pos, _) = p.getBasePositionAndOrientation(body.bodyId)
rayFrom = body.lasers[i].laserRayBatchFrom
rayTo = body.lasers[i].laserRayBatchTo
newFrom = []
newTo = []
for k in range(len(rayFrom)):
newFrom.append([
rayFrom[k][0] + pos[0], rayFrom[k][1] + pos[1],
rayFrom[k][2] + pos[2]
])
newTo.append([
rayTo[k][0] + pos[0], rayTo[k][1] + pos[1],
rayTo[k][2] + pos[2]
])
results = p.rayTestBatch(newFrom, newTo)
for j in range(len(rayFrom)):
hitObjectUid = results[j][0]
if (hitObjectUid < 0):
p.addUserDebugLine(newFrom[j],
newTo[j],
lineColorRGB=[1, 0, 0],
lifeTime=0.1)
else:
hitPosition = results[j][3]
p.addUserDebugLine(newFrom[j],
hitPosition,
lineColorRGB=[0, 1, 0],
lifeTime=0.1)
def getRangeReading(theta):
"""
Simulate laser range readings
:param theta: current angle
:return:
"""
ranges = []
rayFroms = [lidar_pos for i in range(num_rays)]
rayTos = getBatchRayTo(theta)
rayTestInfo = p.rayTestBatch(rayFroms, rayTos)
for i in range(len(rayTestInfo)):
# compute hit distance
d = float(rayTestInfo[i][2]) * lidar_range
# if no hit
if rayTestInfo[i][0] == -1:
d = lidar_range
ranges.append(d)
# Draw debug rays
# for i in range(len(rayTos)):
# p.addUserDebugLine(lidar_pos, rayTos[i], [1,0,0], lifeTime=0.1)
return ranges
def look(self, dist, viewAngle, maxGap):
"""looks at the environment
by performing a ray test batch centered at the position of the character in the direction its facing
input:
dist: maximum sight distance
viewAngle: the field of view (in degree)
maxGap: the maximum arclength between two adjacent raycasts
output:
hitObjList: a list of nested lists that contain the object IDs that were hit and their positions
"""
viewAngle = math.radians(viewAngle)
center = self.pos
z_rot = self.yaw
numRayCasts = math.ceil((dist) * (viewAngle) / (maxGap + 0.0)) + 1
startPosList = [center] * numRayCasts
endPosList = []
viewAngle = maxGap * (numRayCasts - 1) / (
dist + 0.0
) # because numRayCasts is rounded up, original view angle has likely changed
angle = z_rot - viewAngle / 2
for i in range(numRayCasts):
x = center[0] + math.cos(angle) * dist
y = center[1] + math.sin(angle) * dist
z = maxGap / 2.0
endPosList.append([x, y, z])
angle += viewAngle / (numRayCasts - 1)
rayList = p.rayTestBatch(startPosList,
endPosList,
collisionFilterMask=mv.RAY_MASK)
hitObjList = []
for rayOutput in rayList:
if rayOutput[0] != -1 and rayOutput[0] not in hitObjList:
hitID = rayOutput[0]
objPos = hsm.objIDToObject[hitID].pos
hitObjList.append([hitID, objPos])
return hitObjList
def get_lidar_scan(self, extra_outputs=False):
'''
Obtain lidar scan values for current state.
Returns:
list: Scan values for range and resolution specified in
params dict.
'''
if self.valid_buffer_scan:
if extra_outputs:
return self.last_scan, self.last_extra_scan_output
return self.last_scan
scan_front = None
scan_rear = None
scan_l1, scan_h1 = self.get_lidar_rays(self.lidarLinkId1)
scan_l2, scan_h2 = self.get_lidar_rays(self.lidarLinkId2)
scan_l = np.concatenate((scan_l1, scan_l2))
scan_h = np.concatenate((scan_h1, scan_h2))
t0 = time.time()
scan_r = pb.rayTestBatch(scan_l.tolist(), scan_h.tolist(),
self.clientId)
#print (time.time()-t0)
scan = [x[2] * self.p["sensors"]["lidar"]["range"] for x in scan_r]
scan_front = scan[:len(scan_l1)]
scan_rear = scan[len(scan_l1):]
extra_scan_output = [scan_l, scan_h, len(scan_l1)]
self.last_scan = [scan_front, scan_rear]
self.last_extra_scan_output = extra_scan_output
self.valid_buffer_scan = True
if extra_outputs:
return [scan_front, scan_rear], extra_scan_output
return [scan_front, scan_rear]
def raycast(self,
offset=torch.zeros(3),
directions=torch.tensor([1., 0., 0.]),
body=True,
RANGE=100.0):
directions *= RANGE
if len(directions.shape) == 1:
directions = directions.unsqueeze(0)
if len(offset.shape) == 1:
offset = offset.expand(directions.shape[0], -1)
R = self.get_ori(mat=True)
pos = self.get_pos()
if body:
offset = (R @ offset.T).T
directions = (R @ directions.T).T
start = offset + pos
start = start.expand(directions.shape[0], -1)
end = directions + start
rays = p.rayTestBatch(rayFromPositions=start.tolist(),
rayToPositions=end.tolist(),
physicsClientId=self.sim.id)
ray_dict = {}
ray_dict["object"] = [
self.env.object_dict.get(ray[0], None) for ray in rays
]
ray_dict["pos world"] = torch.stack([
torch.tensor(ray[3]) - offset[i, :] for i, ray in enumerate(rays)
])
ray_dict["pos"] = (R.T @ ray_dict['pos world'].T -
(R.T @ pos)[:, None]).T
for i, obj in enumerate(ray_dict["object"]):
if obj is None:
ray_dict["pos world"][i, :] = torch.zeros(3)
ray_dict["pos"][i, :] = torch.zeros(3)
ray_dict["dist"] = ray_dict['pos'].norm(dim=-1)
return ray_dict
def _do_ray_tracing_and_publish(self):
ray_from_position_list = [
self._position_from_matrix(x[0]) for x in self._laser_rays
]
ray_to_position_list = [
self._position_from_matrix(x[1]) for x in self._laser_rays
]
# for i in range(len(ray_from_position_list)):
# rospy.loginfo("RAY {}: from {} to {}".format(i, ray_from_position_list[i], ray_to_position_list[i]))
ray_tracing_output = pybullet.rayTestBatch(
rayFromPositions=ray_from_position_list,
rayToPositions=ray_to_position_list)
self._msg.header.stamp = rospy.Time.now()
self._msg.ranges = []
for ray in ray_tracing_output:
self._msg.ranges.append(self._range_min +
(self._range_max - self._range_min) *
ray[2] - 0.01)
self._laser_scan_publisher.publish(self._msg)
def step(self, action):
# Apply control action (120 Hz, each 1st step)
if (self.stepCounter % 1 == 0 and self.stepCounter != 0):
self.velocity = action[0]
self.steeringAngle = action[1]
self.force = action[2]
self.act()
# Update camera data (120 Hz, each 1st step)
if self.cameraStatus is True:
if (self.stepCounter % 1 == 0):
self.snapshot, self.snapshotPath, self.nextSnapshotPath = self.takeSnapshot()
# Update lidar data (120 Hz, each 1st step)
if (self.stepCounter % 1 == 0):
hitFractions = np.zeros(self.numRays, dtype=np.int8)
distances = np.zeros(self.numRays, dtype=np.float64)
numThreads = 0
results = p.rayTestBatch(self.rayFrom, self.rayTo, numThreads,
parentObjectUniqueId=self.car, parentLinkIndex=self.lidar_joint)
for i in range(self.numRays):
hitObjectUid = results[i][0]
hitFraction = results[i][2]
hitPosition = results[i][3]
hitFractions[i] = hitFraction
if (hitFraction == 1.):
p.addUserDebugLine(self.rayFrom[i], self.rayTo[i], self.rayMissColor, replaceItemUniqueId=self.rayIds[i],
parentObjectUniqueId=self.car, parentLinkIndex=self.lidar_joint)
distances[i] = self.rayLen
else:
localHitTo = [self.rayFrom[i][0] + hitFraction*(self.rayTo[i][0] - self.rayFrom[i][0]),
self.rayFrom[i][1] + hitFraction *
(self.rayTo[i][1] - self.rayFrom[i][1]),
self.rayFrom[i][2] + hitFraction*(self.rayTo[i][2] - self.rayFrom[i][2])]
p.addUserDebugLine(self.rayFrom[i], localHitTo, self.rayHitColor, replaceItemUniqueId=self.rayIds[i],
parentObjectUniqueId=self.car, parentLinkIndex=self.lidar_joint)
distances[i] = np.linalg.norm(
np.array(localHitTo, dtype=np.float64) - self.rayFrom[i])
# self.observation = hitFractions
distances = distances / self.rayLen
self.observation = distances
p.stepSimulation()
carVelocity = p.getBaseVelocity(self.car)[0] # get linear velocity only
carSpeed = np.linalg.norm(carVelocity)
reward = carSpeed*self.dt
if carSpeed <= 0.3:
self.stuckCounter += 1
#print(self.stuckCounter, carSpeed)
if self.stuckCounter >= self.stopThrehold:
done = True
else:
done = False
if self.storeData is True:
datasetRow = [self.snapshotPath, action, reward, int(done), self.nextSnapshotPath]
self.dataset.writerow(datasetRow)
# Update step counter
self.stepCounter += 1
return self.observation, reward, done, {}
camTarget = [camPos[0]+forwardVec[0]*10,camPos[1]+forwardVec[1]*10,camPos[2]+forwardVec[2]*10]
camUpTarget = [camPos[0]+camUpVec[0],camPos[1]+camUpVec[1],camPos[2]+camUpVec[2]]
viewMat = p.computeViewMatrix(camPos, camTarget, camUpVec)
projMat = camInfo[3]
#p.getCameraImage(320,200,viewMatrix=viewMat,projectionMatrix=projMat, flags=p.ER_NO_SEGMENTATION_MASK, renderer=p.ER_BULLET_HARDWARE_OPENGL)
p.getCameraImage(320,200,viewMatrix=viewMat,projectionMatrix=projMat, renderer=p.ER_BULLET_HARDWARE_OPENGL)
lastTime=nowTime
nowControlTime = time.time()
nowLidarTime = time.time()
#lidar at 20Hz
if (nowLidarTime-lastLidarTime>.3):
#print("Lidar!")
numThreads=0
results = p.rayTestBatch(rayFrom,rayTo,numThreads, parentObjectUniqueId=car, parentLinkIndex=hokuyo_joint)
for i in range (numRays):
hitObjectUid=results[i][0]
hitFraction = results[i][2]
hitPosition = results[i][3]
if (hitFraction==1.):
p.addUserDebugLine(rayFrom[i],rayTo[i], rayMissColor,replaceItemUniqueId=rayIds[i],parentObjectUniqueId=car, parentLinkIndex=hokuyo_joint)
else:
localHitTo = [rayFrom[i][0]+hitFraction*(rayTo[i][0]-rayFrom[i][0]),
rayFrom[i][1]+hitFraction*(rayTo[i][1]-rayFrom[i][1]),
rayFrom[i][2]+hitFraction*(rayTo[i][2]-rayFrom[i][2])]
p.addUserDebugLine(rayFrom[i],localHitTo, rayHitColor,replaceItemUniqueId=rayIds[i],parentObjectUniqueId=car, parentLinkIndex=hokuyo_joint)
lastLidarTime = nowLidarTime
#control at 100Hz
if (nowControlTime-lastControlTime>.01):
if (replaceLines): rayIds.append(p.addUserDebugLine(rayFrom[i], rayTo[i], rayMissColor)) else: rayIds.append(-1) if (not useGui): timingLog = p.startStateLogging(p.STATE_LOGGING_PROFILE_TIMINGS,"rayCastBench.json") numSteps = 10 if (useGui): numSteps = 327680 for i in range (numSteps): p.stepSimulation() for j in range (8): results = p.rayTestBatch(rayFrom,rayTo,j+1) #for i in range (10): # p.removeAllUserDebugItems() if (useGui): if (not replaceLines): p.removeAllUserDebugItems() for i in range (numRays): hitObjectUid=results[i][0]
def get_state(self, collision_links=[]):
# calculate state
# sensor_state = self.robots[0].calc_state()
# sensor_state = np.concatenate((sensor_state, self.get_additional_states()))
sensor_state = self.get_additional_states()
state = OrderedDict()
if 'sensor' in self.output:
state['sensor'] = sensor_state
if 'pointgoal' in self.output:
state['pointgoal'] = sensor_state[:2]
if 'rgb' in self.output:
state['rgb'] = self.simulator.renderer.render_robot_cameras(
modes=('rgb'))[0][:, :, :3]
if 'depth' in self.output:
depth = -self.simulator.renderer.render_robot_cameras(
modes=('3d'))[0][:, :, 2:3]
state['depth'] = depth
if 'normal' in self.output:
state['normal'] = self.simulator.renderer.render_robot_cameras(
modes='normal')
if 'seg' in self.output:
state['seg'] = self.simulator.renderer.render_robot_cameras(
modes='seg')
if 'rgb_filled' in self.output:
with torch.no_grad():
tensor = transforms.ToTensor()(
(state['rgb'] * 255).astype(np.uint8)).cuda()
rgb_filled = self.comp(tensor[None, :, :, :])[0].permute(
1, 2, 0).cpu().numpy()
state['rgb_filled'] = rgb_filled
if 'bump' in self.output:
state[
'bump'] = -1 in collision_links # check collision for baselink, it might vary for different robots
if 'pointgoal' in self.output:
state['pointgoal'] = sensor_state[:2]
if 'scan' in self.output:
assert 'scan_link' in self.robots[
0].parts, "Requested scan but no scan_link"
pose_camera = self.robots[0].parts['scan_link'].get_pose()
n_rays_per_horizontal = 128 # Number of rays along one horizontal scan/slice
n_vertical_beams = 9
angle = np.arange(0, 2 * np.pi,
2 * np.pi / float(n_rays_per_horizontal))
elev_bottom_angle = -30. * np.pi / 180.
elev_top_angle = 10. * np.pi / 180.
elev_angle = np.arange(elev_bottom_angle, elev_top_angle,
(elev_top_angle - elev_bottom_angle) /
float(n_vertical_beams))
orig_offset = np.vstack([
np.vstack([
np.cos(angle),
np.sin(angle),
np.repeat(np.tan(elev_ang), angle.shape)
]).T for elev_ang in elev_angle
])
transform_matrix = quat2mat([
pose_camera[-1], pose_camera[3], pose_camera[4], pose_camera[5]
])
offset = orig_offset.dot(np.linalg.inv(transform_matrix))
pose_camera = pose_camera[None, :3].repeat(n_rays_per_horizontal *
n_vertical_beams,
axis=0)
results = p.rayTestBatch(pose_camera, pose_camera + offset * 30)
hit = np.array([item[0] for item in results])
dist = np.array([item[2] for item in results])
dist[dist >= 1 - 1e-5] = np.nan
dist[dist < 0.1 / 30] = np.nan
dist[hit == self.robots[0].robot_ids[0]] = np.nan
dist[hit == -1] = np.nan
dist *= 30
xyz = dist[:, np.newaxis] * orig_offset
xyz = xyz[np.equal(np.isnan(xyz), False)] # Remove nans
#print(xyz.shape)
xyz = xyz.reshape(xyz.shape[0] // 3, -1)
state['scan'] = xyz
return state
matrix = p.getMatrixFromQuaternion(orientation) matrix = np.reshape(matrix, (3, 3)) src = np.array(position) src = src + np.matmul(matrix,offset) path.calcDist(src) h = 10 rays_src = np.repeat([src], num_rays, axis=0) orn = np.matmul(matrix, rot) #rotates unit vector y to -x rays_end = np.matmul(orn, rays) # unit vector in direction of minitaur rays_end = (rays_end + src[:, None]).T rays_info = p.rayTestBatch(rays_src.tolist(), rays_end.tolist()) b = np.asarray([int(i[0]) for i in rays_info]) for i in range(h-1): rays = np.concatenate(([ray_length*np.sin(angles)], [ray_length*np.cos(angles)], [np.full((num_rays,), i+1)]), axis=0) rays_end = np.matmul(orn, rays) # unit vector in direction of minitaur rays_end = (rays_end + src[:, None]).T rays_info = p.rayTestBatch(rays_src.tolist(), rays_end.tolist()) b = np.vstack((b, np.asarray([int(i[0]) for i in rays_info]))) nth_ray = find_largest_gap(b)
def step(self):
quadruped = self.quadruped
bodyPos, bodyOrn = p.getBasePositionAndOrientation(quadruped)
linearVel, angularVel = p.getBaseVelocity(quadruped)
bodyEuler = p.getEulerFromQuaternion(bodyOrn)
kp = p.readUserDebugParameter(self.IDkp)
kd = p.readUserDebugParameter(self.IDkd)
maxForce = p.readUserDebugParameter(self.IDmaxForce)
self.handleCamera(bodyPos, bodyOrn)
self.addInfoText(bodyPos, bodyEuler, linearVel, angularVel)
if self.checkSimulationReset(bodyOrn):
return False
# Calculate Angles with the input of FeetPos,BodyRotation and BodyPosition
angles = self.kin.calcIK(self.Lp, self.rot, self.pos)
for lx, leg in enumerate(
['front_left', 'front_right', 'rear_left', 'rear_right']):
for px, part in enumerate(['shoulder', 'leg', 'foot']):
j = self.jointNameToId[leg + "_" + part]
p.setJointMotorControl2(bodyIndex=quadruped,
jointIndex=j,
controlMode=p.POSITION_CONTROL,
targetPosition=angles[lx][px] *
self.dirs[lx][px],
positionGain=kp,
velocityGain=kd,
force=maxForce)
nowLidarTime = time.time()
if (nowLidarTime - self.lastLidarTime > .2):
numThreads = 0
results = p.rayTestBatch(self.rayFrom,
self.rayTo,
numThreads,
parentObjectUniqueId=quadruped,
parentLinkIndex=0)
for i in range(self.numRays):
hitObjectUid = results[i][0]
hitFraction = results[i][2]
hitPosition = results[i][3]
if (hitFraction == 1.):
if self.debugLidar:
p.addUserDebugLine(self.rayFrom[i],
self.rayTo[i],
self.rayMissColor,
replaceItemUniqueId=self.rayIds[i],
parentObjectUniqueId=quadruped,
parentLinkIndex=0)
else:
localHitTo = [
self.rayFrom[i][0] + hitFraction *
(self.rayTo[i][0] - self.rayFrom[i][0]),
self.rayFrom[i][1] + hitFraction *
(self.rayTo[i][1] - self.rayFrom[i][1]),
self.rayFrom[i][2] + hitFraction *
(self.rayTo[i][2] - self.rayFrom[i][2])
]
dis = math.sqrt(localHitTo[0]**2 + localHitTo[1]**2 +
localHitTo[2]**2)
if self.debugLidar:
p.addUserDebugLine(self.rayFrom[i],
localHitTo,
self.rayHitColor,
replaceItemUniqueId=self.rayIds[i],
parentObjectUniqueId=quadruped,
parentLinkIndex=0)
lastLidarTime = nowLidarTime
# let the Simulator do the rest
if (self.useRealTime):
self.t = time.time() - self.ref_time
else:
self.t = self.t + self.fixedTimeStep
if (self.useRealTime == False):
p.stepSimulation()
time.sleep(self.fixedTimeStep)