-
Notifications
You must be signed in to change notification settings - Fork 1
/
Simulation.py
165 lines (139 loc) · 6.68 KB
/
Simulation.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
import numpy as np
from AircraftModel import *
from Filters import *
from WildfireModels import *
## Wildfire aircraft settings
LAMDA1 = 5.0 # Tuning parameter for reward function
EKF_MODE = True # Observation filter type, True for EKF, False for PF
NUM_UAVS = 2 # Number of UAVs to simulate
'''
Class representing the simulation environment
Has attributes for wildfire simulation, aircraft, and filters
'''
class Sim(object):
def __init__(self,dt,dti,numUAVs = NUM_UAVS, ekf_mode=EKF_MODE,seed=None,windx=-2.,windy=-2.):
self.fireModel = FireModel(seed=seed,windx=windx,windy=windy)
self.numUAVs = numUAVs
self.aircraft = [Aircraft(dt,dti) for i in range(self.numUAVs)]
self.ekf_mode=ekf_mode
if not self.ekf_mode:
self.PF = PF()
self.ekf = independentEKF()
self.reset()
'''
Reset fire model, aircraft, and filters
'''
def reset(self):
self.fireModel.reset()
[ac.reset() for ac in self.aircraft]
self.ekf.reset(self.fireModel.startLocation1)
if not self.ekf_mode:
self.PF.reset()
self.PF_resample_counter=-10
self.seenFireFlag = False
'''
Step the fires forward
'''
def stepFires(self):
self.fireModel.step()
if not self.ekf_mode:
self.PF.step()
self.PF_resample_counter+=1
'''
Step the aircraft forward given a list of actions
'''
def stepAircraft(self,actions):
for ac, action in zip(self.aircraft,actions):
ac.step(action)
'''
Check if all of the fire is within the allowed range
'''
def fireInRange(self,margin=2):
burnX,burnY = np.where(self.fireModel.burnMap.reshape((100,100))==1)
return min(burnX)>=margin and min(burnY)>=margin and max(burnX)<=99-margin and max(burnY)<=99-margin
'''
Updates the filters with observations and computes the reward
'''
def updateEKFandReward(self):
self.ekf.stepTimeMap()
rewards = []
if not self.ekf_mode and self.PF_resample_counter==20:
self.PF.resample()
self.PF_resample_counter=0
for acInd, ac in enumerate(self.aircraft):
## Update EKF given sensor information
sensors = ac.getSensor(self.fireModel.burnMap.reshape((100,100)))
self.ekf.update(sensors)
if not self.ekf_mode:
self.PF.update(sensors)
if not self.seenFireFlag and self.PF_resample_counter<0 and np.sum(self.ekf.seenWildfire)>180:
self.PF_resample_counter=0
self.seenFireFlag=True
# Get Rewards
rewards+= [self.getReward(acInd)]
return rewards
'''
Compute the reward for an aircraft given state of wildfire and its position
The EKF tracks wildfire that has been seen, used to penalize aircraft for flying over seen wildfire
'''
def getReward(self,acInd):
# Penalty for being too close to wildfire
x,y,_,_ = self.aircraft[acInd].state
wildfirePenalty = 0
for i in range(int(x)-4,int(x)+5):
for j in range(int(y)-4,int(y)+5):
if i>=0 and i<100 and j>=0 and j<100 and (x-i)**2+(y-j)<=16:
wildfirePenalty -= LAMDA1 * self.ekf.seenWildfire[j,i]
#Penalty for closeness to intruder
relState = getRelativeState(self.aircraft[0].state,self.aircraft[1].state)
r = relState[1]
proximityPenalty = -1.5*np.exp(-r/10.0)
# Reward for seeing new wildfire
currentWildfireThresh = self.ekf.thresholdMu()
wildfireReward = np.sum((currentWildfireThresh==1) & (self.ekf.seenWildfire==0))
self.ekf.seenWildfire[currentWildfireThresh==1]=1
return proximityPenalty + wildfireReward + wildfirePenalty
'''
This function computes the inputs to the network
The relative belief map, time map, and seen maps are concatenated to form a 3-channel image, and the relative state of the other aircraft is also used
'''
def getBeliefAndState(self, acInd,acInd2):
x,y,th,phi = self.aircraft[acInd].state
rMat = np.sqrt((self.fireModel.gridX.reshape((100,100))-50)**2+(self.fireModel.gridY.reshape((100,100))-50)**2)
thMat = np.arctan2(self.fireModel.gridY.reshape((100,100))-50,self.fireModel.gridX.reshape((100,100))-50)
xRel = np.round(x+rMat*np.cos(th+thMat)).astype(int)
yRel = np.round(y+rMat*np.sin(th+thMat)).astype(int)
xRel2 = np.where((xRel>=0) & (xRel < 100),xRel,0)
yRel2 = np.where((yRel>=0) & (yRel < 100),yRel,0)
xRel3 = np.where((xRel>=-100) & (xRel < 200),xRel,0)
yRel3 = np.where((yRel>=-100) & (yRel < 200),yRel,0)
relTimeMap = np.where((xRel>-100) & (xRel<200) & (yRel>-100) & (yRel<200),self.ekf.timeMap[yRel3+100,xRel3+100],1)
if self.ekf_mode:
relBeliefMap = np.where((xRel>=0) & (xRel<100) & (yRel>=0) & (yRel<100),self.ekf.mu[yRel2,xRel2],0)
relBeliefMapSeen = np.where((xRel>=0) & (xRel<100) & (yRel>=0) & (yRel<100),self.ekf.seenWildfire[yRel2,xRel2],0)
else:
burnThresh = self.PF.getMeanBurnThresh().reshape((100,100))
burnedMap = self.PF.getBurned().reshape((100,100))
relBeliefMap = np.where((xRel>=0) & (xRel<100) & (yRel>=0) & (yRel<100),burnThresh[yRel2,xRel2],0)
relBeliefMapSeen = np.where((xRel>=0) & (xRel<100) & (yRel>=0) & (yRel<100),burnedMap[yRel2,xRel2],0)
relState = getRelativeState(self.aircraft[acInd].state,self.aircraft[acInd2].state)
return [np.stack([relBeliefMap,relTimeMap,relBeliefMapSeen],axis=2), relState]
'''
Computes vertices of sensor region on ground using a Convex Hull
'''
def getSensorVerts(self):
sensorVerts = {}
for acInd, ac in enumerate(self.aircraft):
sensorVerts[acInd] = []
sensorTuples = ac.getSensor(self.fireModel.burnMap.reshape((100,100)))
for sensor in sensorTuples:
points, obs = sensor
inds = np.where((points[0]>0) & (points[0]<99) & (points[1]>0) & (points[1]<99))[0]
try:
hull = ConvexHull(points[:,inds].T)
if hull is not None:
vert = np.concatenate((hull.vertices,np.array([hull.vertices[0]])))
sensorVerts[acInd] += [hull.points[vert]]
except:
sensorVerts[acInd] += [np.array([])]
return sensorVerts