-
Notifications
You must be signed in to change notification settings - Fork 0
/
DirectPathQRCost.py
185 lines (140 loc) · 5.94 KB
/
DirectPathQRCost.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
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
import os
import sys
module_path = os.path.abspath(os.path.join('../ilqr'))
if module_path not in sys.path:
sys.path.append(module_path)
import numpy as np
import matplotlib.pyplot as plt
import decimal
import copy
from ilqr import iLQR
from ilqr.cost import Cost
from ilqr.cost import QRCost
from ilqr.cost import PathQRCost, AutoDiffCost, FiniteDiffCost
from ilqr.dynamics import constrain
from ilqr.examples.pendulum import InvertedPendulumDynamics
from ilqr.dynamics import BatchAutoDiffDynamics, tensor_constrain
from LegiblePathQRCost import LegiblePathQRCost
from scipy.optimize import approx_fprime
import utility_legibility as legib
import utility_environ_descrip as resto
import pipeline_generate_paths as pipeline
class DirectPathQRCost(LegiblePathQRCost):
FLAG_DEBUG_J = True
"""Quadratic Regulator Instantaneous Cost for trajectory following."""
def __init__(self, Q, R, x_path, u_path, start, target_goal, goals, N, dt, Q_terminal=None):
"""Constructs a QRCost.
Args:
Q: Quadratic state cost matrix [state_size, state_size].
R: Quadratic control cost matrix [action_size, action_size].
x_path: Goal state path [N+1, state_size].
u_path: Goal control path [N, action_size].
Q_terminal: Terminal quadratic state cost matrix
[state_size, state_size].
"""
self.Q = np.array(Q)
self.R = np.array(R)
self.x_path = np.array(x_path)
self.start = np.array(start)
self.goals = goals
self.target_goal = target_goal
self.N = N
self.dt = dt
# Create a restaurant object for using those utilities, functions, and print functions
# dim gives the dimensions of the restaurant
self.restaurant = resto.Restaurant(resto.TYPE_CUSTOM, tables=[], goals=goals, start=start, observers=[], dim=None)
state_size = self.Q.shape[0]
action_size = self.R.shape[0]
path_length = self.x_path.shape[0]
x_eps = .01 #05
u_eps = .01 #05
# self._x_eps_hess = np.sqrt(self._x_eps)
# self._u_eps_hess = np.sqrt(self._u_eps)
self._state_size = state_size
self._action_size = action_size
if Q_terminal is None:
self.Q_terminal = self.Q
else:
self.Q_terminal = np.array(Q_terminal)
if u_path is None:
self.u_path = np.zeros(path_length - 1, action_size)
else:
self.u_path = np.array(u_path)
assert self.Q.shape == self.Q_terminal.shape, "Q & Q_terminal mismatch"
assert self.Q.shape[0] == self.Q.shape[1], "Q must be square"
assert self.R.shape[0] == self.R.shape[1], "R must be square"
assert state_size == self.x_path.shape[1], "Q & x_path mismatch"
assert action_size == self.u_path.shape[1], "R & u_path mismatch"
assert path_length == self.u_path.shape[0] + 1, \
"x_path must be 1 longer than u_path"
# Precompute some common constants.
self._Q_plus_Q_T = self.Q + self.Q.T
self._R_plus_R_T = self.R + self.R.T
self._Q_plus_Q_T_terminal = self.Q_terminal + self.Q_terminal.T
LegiblePathQRCost.__init__(
self,
Q, R, x_path, u_path, start, target_goal, goals, N, dt, Q_terminal=Q_terminal
)
# How far away is the final step in the path from the goal?
def term_cost(self, x, i):
start = self.start
goal1 = self.target_goal
Qf = self.Q_terminal
R = self.R
x_diff = x - self.x_path[i]
squared_x_cost = .5 * x_diff.T.dot(Qf).dot(x_diff)
terminal_cost = squared_x_cost
# We want to value this highly enough that we don't not end at the goal
terminal_coeff = 1000.0
terminal_cost = terminal_cost * terminal_coeff
return terminal_cost
# original version for plain path following
def l(self, x, u, i, terminal=False):
"""Instantaneous cost function.
Args:
x: Current state [state_size].
u: Current control [action_size]. None if terminal.
i: Current time step.
terminal: Compute terminal cost. Default: False.
Returns:
Instantaneous cost (scalar).
"""
Q = self.Q_terminal if terminal else self.Q
R = self.R
x_diff = x - self.x_path[i]
squared_x_cost = x_diff.T.dot(Q).dot(x_diff)
if terminal:
return squared_x_cost
u_diff = u - self.u_path[i]
return squared_x_cost + u_diff.T.dot(R).dot(u_diff)
# Removed version that overcounted
# start = self.start
# goal = self.target_goal
# if terminal:
# return self.term_cost(x, i)
# else:
# # difference between this step and the end
# term_cost = self.term_cost(x, i)
# stage_costs = self.get_total_stage_cost(start, goal, x, u, i, terminal)
# print("STAGE,\t TERM")
# print(stage_costs, term_cost)
# total = term_cost + stage_costs
# return float(total)
def get_total_stage_cost(self, start, goal, x, u, i, terminal):
N = self.N
R = self.R
stage_costs = 0.0
# print("u = " + str(u))
# print("Getting stage cost")
for j in range(i):
u_diff = u - self.u_path[j]
x_diff = x - self.x_path[j]
# print("at " + str(j) + "u_diff = " + str(u_diff))
# print(u_diff.T.dot(R).dot(u_diff))
# stage_costs += self.michelle_stage_cost(start, goal, x, u, j, terminal)
stage_costs += u_diff.T.dot(R).dot(u_diff)
stage_costs += x_diff.T.dot(R).dot(x_diff)
# stage_cost(x, u, j, terminal) #
# stage_costs = stage_costs + self.goal_efficiency_through_point_relative(start, goal, x, terminal)
print("total stage cost " + str(stage_costs))
return stage_costs