XTDrone/coordination/formation_demo/follower_consensus.py

408 lines
20 KiB
Python

#!/usr/bin/python
# -*- coding: UTF-8 -*-
### This code is about the distributed formation control of consensus protocol with a certain
### Laplacian matrix and the formation transformation based on a task allocation algorithm——
### KM for the shorest distances of all the UAVs to achieve the new pattern.
### For more information of these two algorithms, please see the latest paper on https://arxiv.org/abs/2005.01125
import rospy
from geometry_msgs.msg import Twist, Vector3, PoseStamped, TwistStamped
from std_msgs.msg import String
import sys
import heapq
import copy
# formation patterns
if sys.argv[3] == '6':
from formation_dict import formation_dict_6 as formation_dict
elif sys.argv[3] == '9':
from formation_dict import formation_dict_9 as formation_dict
elif sys.argv[3] == '18':
from formation_dict import formation_dict_18 as formation_dict
import numpy
import Queue
class Follower:
def __init__(self, uav_type, uav_id, uav_num):
self.hover = "HOVER"
self.offboard = "OFFBOARD"
self.uav_type = uav_type
self.id = uav_id
self.uav_num = uav_num
self.f = 30 # control/communication rate
self.local_pose = PoseStamped()
self.local_pose_queue = Queue.Queue(self.f / 10)
for i in range(self.f / 10):
self.local_pose_queue.put(PoseStamped())
self.local_velocity = TwistStamped()
self.cmd_vel_enu = Twist()
self.avoid_vel = Vector3()
self.following_switch = False # determine whether the formation pattern is required to be changed
self.arrive_print = True # determine whether the target position has been reached
self.following_ids = [] # followers of this uav
self.formation_config = 'waiting'
self.following_count = 0 # the number of followers of this uav
self.Kp = 0.2
self.velxy_max = 1 # 0.8
self.velz_max = 1
self.following_local_pose = [PoseStamped() for i in range(
self.uav_num)] # local position of other uavs, and only the position of followers of this uav is not zero
self.following_local_pose_sub = [[] for i in range(
self.uav_num)]
self.arrive_count = 0
self.local_pose_sub = rospy.Subscriber(self.uav_type + '_' + str(self.id) + "/mavros/local_position/pose",
PoseStamped, self.local_pose_callback)
self.avoid_vel_sub = rospy.Subscriber("/xtdrone/" + self.uav_type + '_' + str(self.id) + "/avoid_vel", Vector3,
self.avoid_vel_callback)
self.formation_switch_sub = rospy.Subscriber("/xtdrone/formation_switch", String,
self.formation_switch_callback)
self.vel_enu_pub = rospy.Publisher('/xtdrone/' + self.uav_type + '_' + str(self.id) + '/cmd_vel_enu', Twist,
queue_size=10)
self.info_pub = rospy.Publisher('/xtdrone/' + self.uav_type + '_' + str(self.id) + '/info', String,
queue_size=10)
self.cmd_pub = rospy.Publisher('/xtdrone/' + self.uav_type + '_' + str(self.id) + '/cmd', String, queue_size=10)
for i in range(self.uav_num):
self.following_local_pose_sub[i] = rospy.Subscriber(
self.uav_type + '_' + str(i) + "/mavros/local_position/pose", PoseStamped,
self.following_local_pose_callback, i)
self.first_formation = True
self.orig_formation = None
self.new_formation = None
def local_pose_callback(self, msg):
self.local_pose = copy.deepcopy(msg)
pose_comparison = self.local_pose_queue.get()
self.local_pose_queue.put(self.local_pose)
comparison = (self.local_pose.pose.position.x - pose_comparison.pose.position.x) ** 2 + (
self.local_pose.pose.position.y - pose_comparison.pose.position.y) ** 2 + (
self.local_pose.pose.position.z - pose_comparison.pose.position.z) ** 2
if comparison < float(
self.velxy_max ** 2 + self.velxy_max ** 2 + self.velz_max ** 2) / 1e5: # if the target position is reached, arrive_count ++1
self.arrive_count += 1
else:
self.arrive_count = 0
def following_local_pose_callback(self, msg, id):
self.following_local_pose[id] = copy.deepcopy(msg)
# the order of changing the formation pattern
def formation_switch_callback(self, msg):
if not self.formation_config == msg.data:
self.following_switch = True
self.formation_config = msg.data
def avoid_vel_callback(self, msg):
self.avoid_vel = copy.deepcopy(msg)
def loop(self):
rospy.init_node('follower' + str(self.id - 1))
rate = rospy.Rate(self.f)
while not rospy.is_shutdown():
if self.arrive_count > 2000 and self.arrive_print:
print("Follower" + str(self.id - 1) + ":Arrived")
self.arrive_print = False
if self.following_switch:
self.following_switch = False
self.arrive_print = True
self.arrive_count = 0
for i in range(self.f / 10):
self.cmd_pub.publish(self.offboard)
self.info_pub.publish("Received")
print("Follower" + str(self.id - 1) + ": Switch to Formation " + self.formation_config)
# The Laplacian matrix is invarible in this code, and you can change it if necessary.
if self.formation_config == 'waiting':
self.L_matrix = self.get_L_central_matrix()
else:
# Change from the original pattern to the first pattern without KM.
if self.first_formation:
self.first_formation = False
self.orig_formation = formation_dict[self.formation_config]
self.L_matrix = self.get_L_central_matrix()
else:
self.adj_matrix = self.build_graph(self.orig_formation,
formation_dict[self.formation_config])
# These variables are determined for KM algorithm, see examples of KM algorithm on Github.
self.label_left = numpy.max(self.adj_matrix, axis=1) # init label for the left
self.label_right = numpy.array([0] * (self.uav_num - 1)) # init label for the right set
self.match_right = numpy.array([-1] * (self.uav_num - 1))
self.visit_left = numpy.array([0] * (self.uav_num - 1))
self.visit_right = numpy.array([0] * (self.uav_num - 1))
self.slack_right = numpy.array([100] * (self.uav_num - 1))
self.change_id = self.KM()
# Get a new formation pattern of UAVs based on KM.
# self.L_matrix = self.get_L_matrix(self.orig_formation)
self.new_formation = self.get_new_formation(self.change_id,
formation_dict[self.formation_config])
self.L_matrix = self.get_L_matrix(self.new_formation)
# self.L_matrix = self.get_L_central_matrix()
self.orig_formation = self.new_formation
if self.id == 3:
print(self.L_matrix)
# Get the followers of this uav based on the Laplacian matrix, and update the position of the followers.
self.following_ids = numpy.argwhere(self.L_matrix[self.id, :] == 1)
self.following_count = 0
# for i in range(self.uav_num):
# if not self.following_local_pose_sub[i] == None:
# self.following_local_pose_sub[i].unregister()
self.following_count = len(self.following_ids)
print(self.id, "UAV's following count: ",self.following_count)
self.cmd_vel_enu.linear = Vector3(0, 0, 0)
input_vel = Vector3(0, 0, 0)
# Code of the consensus protocol, see details on the paper.
for following_id in self.following_ids:
if following_id[0] == 0:
input_vel.x += self.following_local_pose[
following_id[0]].pose.position.x - self.local_pose.pose.position.x + \
self.new_formation[0, self.id - 1]
input_vel.y += self.following_local_pose[
following_id[0]].pose.position.y - self.local_pose.pose.position.y + \
self.new_formation[1, self.id - 1]
input_vel.z += self.following_local_pose[
following_id[0]].pose.position.z - self.local_pose.pose.position.z + \
self.new_formation[2, self.id - 1]
else:
input_vel.x += self.following_local_pose[
following_id[0]].pose.position.x - self.local_pose.pose.position.x + \
self.new_formation[0, self.id - 1] - self.new_formation[0, following_id[0] - 1]
input_vel.y += self.following_local_pose[
following_id[0]].pose.position.y - self.local_pose.pose.position.y + \
self.new_formation[1, self.id - 1] - self.new_formation[1, following_id[0] - 1]
input_vel.z += self.following_local_pose[
following_id[0]].pose.position.z - self.local_pose.pose.position.z + \
self.new_formation[2, self.id - 1] - self.new_formation[2, following_id[0] - 1]
if self.id == 4:
print(following_id[0],"'s pose: ", self.following_local_pose[
following_id[0]].pose.position)
print(self.id , "'s my pose: ", self.local_pose.pose.position)
print(input_vel)
omega = self.Kp/self.following_count
self.cmd_vel_enu.linear.x = omega * input_vel.x + self.avoid_vel.x
self.cmd_vel_enu.linear.y = omega * input_vel.y + self.avoid_vel.y
self.cmd_vel_enu.linear.z = omega * input_vel.z + self.avoid_vel.z
if self.cmd_vel_enu.linear.x > self.velxy_max:
self.cmd_vel_enu.linear.x = self.velxy_max
elif self.cmd_vel_enu.linear.x < - self.velxy_max:
self.cmd_vel_enu.linear.x = - self.velxy_max
if self.cmd_vel_enu.linear.y > self.velxy_max:
self.cmd_vel_enu.linear.y = self.velxy_max
elif self.cmd_vel_enu.linear.y < - self.velxy_max:
self.cmd_vel_enu.linear.y = - self.velxy_max
if self.cmd_vel_enu.linear.z > self.velz_max:
self.cmd_vel_enu.linear.z = self.velz_max
elif self.cmd_vel_enu.linear.z < - self.velz_max:
self.cmd_vel_enu.linear.z = - self.velz_max
if not self.formation_config == 'waiting':
self.vel_enu_pub.publish(self.cmd_vel_enu)
if (self.cmd_vel_enu.linear.x) ** 2 + (self.cmd_vel_enu.linear.y) ** 2 + (
self.cmd_vel_enu.linear.z) ** 2 < 0.2:
self.arrive_count += 1
else:
self.arrive_count = 0
rate.sleep()
# 'build_graph', 'find_path' and 'KM' functions are all determined for KM algorithm.
# A graph of UAVs is established based on distances between them in 'build_graph' function.
def build_graph(self, orig_formation, change_formation):
distance = [[0 for i in range(self.uav_num - 1)] for j in range(self.uav_num - 1)]
for i in range(self.uav_num - 1):
for j in range(self.uav_num - 1):
distance[i][j] = numpy.linalg.norm(orig_formation[:, i] - change_formation[:, j])
distance[i][j] = int(50 - distance[i][j])
return distance
# Determine whether a path has been found.
def find_path(self, i):
self.visit_left[i] = True
for j, match_weight in enumerate(self.adj_matrix[i], start=0):
if self.visit_right[j]:
continue
gap = self.label_left[i] + self.label_right[j] - match_weight
if gap == 0:
self.visit_right[j] = True
if self.match_right[j] == -1 or self.find_path(self.match_right[j]):
self.match_right[j] = i
return True
else:
self.slack_right[j] = min(gap, self.slack_right[j])
return False
# Main body of KM algorithm.
def KM(self):
for i in range(self.uav_num - 1):
self.slack_right = numpy.array([100] * (self.uav_num - 1))
while True:
self.visit_left = numpy.array([0] * (self.uav_num - 1))
self.visit_right = numpy.array([0] * (self.uav_num - 1))
if self.find_path(i):
break
d = numpy.inf
for j, slack in enumerate(self.slack_right):
if not self.visit_right[j]:
d = min(d, slack)
for k in range(self.uav_num - 1):
if self.visit_left[k]:
self.label_left[k] -= d
if self.visit_right[k]:
self.label_right[k] += d
else:
self.slack_right[k] -= d
return self.match_right
# The formation patterns designed in the formation dictionaries are random (the old ones),
# and a new formation pattern based on the distances of UAVs of the current pattern is designed as follows.
# Note that only the desired position of each UAV has changed, while the form of the new pattern is the same as the one in the dictionary.
def get_new_formation(self, change_id, change_formation):
new_formation = numpy.zeros((3, self.uav_num - 1))
position = numpy.zeros((3, self.uav_num - 1))
change_id = [i + 1 for i in change_id]
for i in range(0, self.uav_num - 1):
position[:, i] = change_formation[:, i]
for i in range(0, self.uav_num - 1):
for j in range(0, self.uav_num - 1):
if (j + 1) == change_id[i]:
new_formation[:, i] = position[:, j]
return new_formation
# Laplacian matrix
def get_L_matrix(self, rel_posi):
c_num = int((self.uav_num) / 2)
min_num_index_list = [0] * c_num
comm = [[] for i in range(self.uav_num)]
w = numpy.ones((self.uav_num, self.uav_num)) * 0
nodes_next = []
node_flag = [self.uav_num - 1]
node_mid_flag = []
rel_d = [0] * (self.uav_num - 1)
for i in range(0, self.uav_num - 1):
rel_d[i] = pow(rel_posi[0][i], 2) + pow(rel_posi[1][i], 2) + pow(rel_posi[2][i], 2)
c = numpy.copy(rel_d)
c.sort()
count = 0
for j in range(0, c_num):
for i in range(0, self.uav_num - 1):
if rel_d[i] == c[j]:
if not i in node_mid_flag:
min_num_index_list[count] = i
node_mid_flag.append(i)
count = count + 1
if count == c_num:
break
if count == c_num:
break
for j in range(0, c_num):
nodes_next.append(min_num_index_list[j])
comm[self.uav_num - 1].append(min_num_index_list[j])
size_ = len(node_flag)
while (nodes_next != []) and (size_ < (self.uav_num - 1)):
next_node = nodes_next[0]
nodes_next = nodes_next[1:]
min_num_index_list = [0] * c_num
node_mid_flag = []
rel_d = [0] * (self.uav_num - 1)
for i in range(0, self.uav_num - 1):
if i == next_node or i in node_flag:
rel_d[i] = 2000
else:
rel_d[i] = pow((rel_posi[0][i] - rel_posi[0][next_node]), 2) + pow(
(rel_posi[1][i] - rel_posi[1][next_node]), 2) + pow((rel_posi[2][i] - rel_posi[2][next_node]),
2)
c = numpy.copy(rel_d)
c.sort()
count = 0
for j in range(0, c_num):
for i in range(0, self.uav_num - 1):
if rel_d[i] == c[j]:
if not i in node_mid_flag:
min_num_index_list[count] = i
node_mid_flag.append(i)
count = count + 1
if count == c_num:
break
if count == c_num:
break
node_flag.append(next_node)
size_ = len(node_flag)
for j in range(0, c_num):
if min_num_index_list[j] in node_flag:
nodes_next = nodes_next
else:
if min_num_index_list[j] in nodes_next:
nodes_next = nodes_next
else:
nodes_next.append(min_num_index_list[j])
comm[next_node].append(min_num_index_list[j])
for i in range(0, self.uav_num):
for j in range(0, self.uav_num - 1):
if i == 0:
if j in comm[self.uav_num - 1]:
w[j + 1][i] = 1
else:
w[j + 1][i] = 0
else:
if j in comm[i - 1] and i < (j+1):
w[j + 1][i] = 1
else:
w[j + 1][i] = 0
for i in range(1, self.uav_num): # 防止某个无人机掉队
if sum(w[i]) == 0:
w[i][0] = 1
L = w
for i in range(0, self.uav_num):
L[i][i] = -sum(w[i])
return L
def get_L_central_matrix(self):
L = numpy.zeros((self.uav_num, self.uav_num))
for i in range(1, self.uav_num):
L[i][0] = 1
L[i][i] = -1
# L = numpy.array([[0,0,0,0,0,0],[1,-1,0,0,0,0],[1,1,-2,0,0,0],[1,1,1,-3,0,0],[1,1,1,1,-4,0],[1,1,1,1,1,-5]])
# L=numpy.array([[-0., 0. , 0. , 0. , 0. , 0. , 0. , 0. , 0.], [ 1. ,-1., 0., 0., 0., 0., 0., 0., 0.], [ 1., 1., -2., 0., 0., 0. , 0. , 0. , 0.],
# [ 1., 1., 1., -3., 0., 0., 0., 0., 0.],[ 1., 1., 1., 1., -4., 0., 0., 0., 0.],
# [0., 1., 1., 0., 1., - 3., 0., 0., 0.],[ 0., 0., 1., 1., 1., 1., -4., 0., 0.],
# [0., 0., 0., 1., 1., 1., 1., - 4., 0.], [ 0., 0., 0., 1., 1., 1., 1., 1., -5.]])
return L
if __name__ == '__main__':
follower = Follower(sys.argv[1], int(sys.argv[2]), int(sys.argv[3]))
follower.loop()