class Navigator(object):
"""docstring for Navigator"""
def __init__(self, start_pos, target_list=[]):
"""
Various parameters we use to make decisions on
"""
# The start position we want to hit before we follow targets
self.start_pos = start_pos
# List of targets to follow
self.target_list = target_list
# nav_queue is used to contain a list of lists of coordinates to follow
self.nav_queue = list()
# The distance required of closeness to current coordinate for assuming it to be reached
self.coord_dist_cutoff = 0.6
# What the current intermediate goal is
self.curr_target = (0.0, 0.0)
# Gains for calculating linear and angular speed
# ref: A Stable Target-Tracking Control for Unicycle Mobile Robots Lee, S. et al.
# Linear velocity depends only on K1, angular depends on K1 and K2
self.K1_default = 1.5
self.K2_default = -5.0
self.K1_slow = 0.5
self.K2_slow = -5.0
self.K1 = self.K1_default
self.K2 = self.K2_default
# Set up a markerarray to show the planned path to take
self.planned_markers = MarkerPlacer("rviz_planned_path", "map", 4000)
# Make the marker size a bit bigger so that they are easier to see
self.planned_markers.set_scale(0.12, 0.12, 0.12)
# Set them to be spheres
self.planned_markers.set_type(Marker.SPHERE)
# Pick a random colour since we might have multiple planned paths and want
# to differentiate between them
self.planned_markers.random_color()
# Set up a markerarray to show the actual path taken
self.taken_markers = MarkerPlacer("rviz_taken_path", "map", 4000)
# Make the marker size a bit bigger so that they are easier to see
self.taken_markers.set_scale(0.12, 0.12, 0.12)
# Set them to be spheres
self.taken_markers.set_type(Marker.SPHERE)
# Pick a random colour since we might want to follow multiple paths and
# want to differentiate between them
self.taken_markers.random_color()
# Set up a markerarray to place markers when the goal is reached
self.goal_markers = MarkerPlacer("rviz_goals", "map", 40)
# Make the marker size a bit bigger so that they are easier to see
self.goal_markers.set_scale(0.10, 0.10, 1.35)
# Set them to be spheres
self.goal_markers.set_type(Marker.CYLINDER)
# Set colour to pink
self.goal_markers.set_color(1.0, 0.0, 1.0, 1.0)
# Add a marker for the assumed pose that the navigator is acting on
self.pose_marker = MarkerPlacer("/rviz_nav_pose", "/map")
self.pose_marker.set_color(1.0, 1.0, 0.0, 1.0)
self.pose_marker.set_scale(1.25, 0.25, 0.25)
# Set up a container for current pose of robot. This pose is the fusion of
# data received from the particle filter and odometry updates.
# We store yaw in a separate variable since this will reduce the number of
# transformations is required from going quat -> yaw -> quat since we will
# use the yaw directly for navigation purposes
self.pose = Pose()
self.yaw = 0.0
# Subscribe to the real pose of the robot in order to drop markers along the followed path
self.real_pose = PoseStamped()
rospy.Subscriber("/real_pose", PoseStamped, self.get_real_pose)
# Set up a subscriber to particle filter pose
self.particle_pose = PoseStamped()
rospy.Subscriber("/particle_pose", PoseStamped, self.get_particle_pose)
# At this point we kind-of want to wait until we have a particle update
# before we start throwing odometry updates at the uninitialised position.
rospy.wait_for_message("/particle_pose", Odometry)
# Set up a subscriber for (noisy) odometry updates
# First wait for an Odometry message in order to get a time stamp which
# will be used to update the robots position from the odometry data
rospy.wait_for_message("/odom", Odometry)
self.last_odom_time = rospy.get_time()
# Then create a container for the odometry and subscribe to the topic.
self.odom = Odometry()
rospy.Subscriber("/odom", Odometry, self.get_odom)
# Initialize the map that will be used for calculating paths and overall navigation
self.m_s = None
self.initialize_map()
# Find path from robot position to start position. For now we're doing
# this in the navigator, but should perhaps be outsources to an external
# module later since it can take a very long time.
self.xy_path = list()
self.xy_pos_path = list()
self.get_path_to_start()
# Add the start pos since the resolution of the planned path is not perfect
self.xy_pos_path.append(start_pos)
# Place markers for the current path to show that we know where we are going (hopefully!)
self.mark_current_path()
# At this point we got: Robot pose (fusion of particle + odom) and a list
# of coordinates we want to follow towards a goal. This should be enough
# navigate!
# Set up a publisher for sending velocity commands to the robot
self.vel = Twist()
self.vel_publisher = rospy.Publisher("/cmd_vel", Twist, queue_size=10)
# Pop off the first item in list as current goal
self.curr_target = self.xy_pos_path.pop(0)
def run(self):
# Here we need to act on the information previously received regarding
# robot pose and list of coordinates to follow.
# Using these parameters we have to figure out:
# 1) When we are close enough to a coordinate to knock it off the list
# 1.1) Is the target the goal? If so -> What to do?
# 2) What the next coordinate is
# 3) The angle towards the next coordinate
# 4) The angular and linear speed we need to set in order to hit the coordinate
# Find closeness to current coordinate
curr_dist = dist((self.pose.position.x, self.pose.position.y), self.curr_target)
if curr_dist < self.coord_dist_cutoff:
# Drop off a marker to show that we have been here
self.taken_markers.place_marker([copy.deepcopy(self.real_pose.pose.position)],
[copy.deepcopy(self.real_pose.pose.orientation)])
# Check if we have arrived at the last point in the list (the ultimate goal)
if len(self.xy_pos_path) == 0:
if curr_dist < 0.2:
# Poop goal marker
self.goal_markers.place_marker([self.real_pose.pose.position],
[self.real_pose.pose.orientation])
# Check if there are more targets to go to
if len(self.target_list) > 0:
# Change the colour for the planned path and path taken to differentiate
# it from the previous path
self.taken_markers.random_color()
self.planned_markers.random_color()
# Find a new target and path
self.xy_path = self.get_next_path()
# Convert the path to xy_pos format
self.convert_current_path()
# Mark it
self.mark_current_path()
# if no more targets go to sleep
else:
rospy.sleep()
# If we have more coordinates to visit pop off the next one!
else:
self.curr_target = self.xy_pos_path.pop(0)
curr_dist = dist((self.pose.position.x, self.pose.position.y), self.curr_target)
# Adjust the speed depending on how close we are to the current goal
if len(self.xy_pos_path) < 5:
self.K1 = self.K1_slow
self.K2 = self.K2_slow
else:
self.K1 = self.K1_default
self.K2 = self.K2_default
# Figure out the angle we want to travel in to reach the current target
# Assumption: This is given by atan2 between the target and the robot?
# This makes sense as it would transpose the origin to the centre of the
# robot, leaving the calculation to a "virtual" coordinate system around
# the robot.
theta = math.atan2(self.curr_target[1] - self.pose.position.y,
self.curr_target[0] - self.pose.position.x)
# Find the difference in theta that should go towards zero when the
# correct heading is reached
d_theta = theta - self.yaw
d_theta = self.normalize(d_theta)
# Find a new set of linear and angular speeds to publish
lin_speed = self.new_speed(curr_dist, d_theta)
ang_speed = self.new_angular_speed(d_theta)
self.vel.linear.x = lin_speed
self.vel.angular.z = ang_speed
self.vel_publisher.publish(self.vel)
def new_speed(self, radius, target_theta):
"""
ref:
A Stable Target-Tracking Control for Unicycle Mobile Robots
Lee, S. et al.
"""
return self.K1 * radius * math.cos(target_theta)
def new_angular_speed(self, target_theta):
"""
ref:
A Stable Target-Tracking Control for Unicycle Mobile Robots
Lee, S. et al.
"""
factor_1 = (-self.K1) * math.sin(target_theta) * math.cos(target_theta)
factor_2 = self.K2 * target_theta
return factor_1 - factor_2
def get_real_pose(self, pose):
self.real_pose = pose
def get_particle_pose(self, particle_pose):
prev_particle_pose = self.particle_pose
self.particle_pose = particle_pose
# Set up a check to see if the new pose is out of whack - Might happen on
# random occasions. If it is we do not want to update the assumed pose but
# rather continue using the estimated pose from previous particle filter +
# odometry updates
# TODO(geir): Implement the check somehow
if True:
self.pose.position = self.particle_pose.pose.position
self.pose.orientation = self.particle_pose.pose.orientation
pitch, roll, yaw = tf.transformations.euler_from_quaternion([self.pose.orientation.x,
self.pose.orientation.y,
self.pose.orientation.z,
self.pose.orientation.w])
self.yaw = yaw
def get_odom(self, odom):
self.odom = odom
# Update current pose with odom data!
# First get the current time and time since last update
time = rospy.get_time()
d_time = time-self.last_odom_time
self.last_odom_time = time
# Get the linear and angular velocities
lin_vel = self.odom.twist.twist.linear.x
ang_vel = self.odom.twist.twist.angular.z
# Calculate change in yaw, x and y
d_theta = ang_vel * d_time
d_x = (lin_vel * math.cos(self.yaw))*d_time
d_y = (lin_vel * math.sin(self.yaw))*d_time
# Update position
self.yaw += d_theta
self.pose.position.x += d_x
self.pose.position.y += d_y
q = tf.transformations.quaternion_from_euler(0.0, 0.0, self.yaw)
self.pose.orientation = Quaternion(*q)
# Publish the marker for this pose
self.pose_marker.place_marker([self.pose.position], [self.pose.orientation])
def initialize_map(self):
# Load a map, down sample it and expand it
self.m_s = MapStorage()
# Down sample in order to reduce graph size
self.m_s.downsample_map_by_half()
self.m_s.downsample_map_by_half()
# Expand in order to avoid collisions
# Previously expanded twice (inside down sampling) => 0.1 res -> 0.4 res
# Three more expansions will leave us with an object previously taking up a
# size of 0.4 m to take up ~4*0.4 -> 1.6. The robot size is 1 metre, so
# this leaves a space of 0.6 for errors before collisions occur.
# This is *MUCH* needed - anything else leads to a boatload of collisions
self.m_s.expand_map()
self.m_s.expand_map()
self.m_s.expand_map()
self.m_s.expand_map()
def get_next_path(self):
# Here we want to find the item in the list which is the closest to the
# current position of the robot.
closest = 999999.9
closest_index = 0
for i in range(len(self.target_list)):
d = dist((self.pose.position.x, self.pose.position.y),
(self.target_list[i][0], self.target_list[i][1]))
if d < closest:
closest = d
closest_index = i
# Get the target that is closest
pos_t = self.target_list.pop(closest_index)
# Find the x,y values for robot position and target
x_t, y_t = self.m_s.xy_from_xy_pos(pos_t[0], pos_t[1])
x_r, y_r = self.m_s.xy_from_xy_pos(self.pose.position.x, self.pose.position.y)
# Search for path
path = astar_search(self.m_s, (x_r, y_r), (x_t, y_t))
return path
def get_path_to_start(self):
# Convert x_pos,y_pos for self.pose and self.start_pos to x,y for use with A* search
x_r, y_r = self.m_s.xy_from_xy_pos(self.pose.position.x, self.pose.position.y)
x_s, y_s = self.m_s.xy_from_xy_pos(self.start_pos[0], self.start_pos[1])
# Find a path from current position to start position
self.xy_path = astar_search(self.m_s, (x_r, y_r), (x_s, y_s))
# Convert the path in to real coordinates
self.convert_current_path()
def convert_current_path(self):
# The path is now in (x,y) format so we want to convert this to a list in
# (x_pos,y_pos) format in order to use it for navigation
self.xy_pos_path = list()
for p in self.xy_path:
x_pos, y_pos = self.m_s.xy_pos_from_xy(p[0], p[1])
self.xy_pos_path.append((x_pos, y_pos))
def mark_current_path(self):
o = Quaternion() # Need for interface
o_list = list()
p_list = list()
for pos in self.xy_pos_path:
o_list.append(o)
p_list.append(Point(pos[0], pos[1], 0.0))
self.planned_markers.place_marker(p_list, o_list)
def normalize(self, theta):
while theta <= math.pi:
theta += 2.0*math.pi
while theta > math.pi:
theta -= 2.0*math.pi
return theta
