-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathddm_argparse.py
175 lines (157 loc) · 8.81 KB
/
ddm_argparse.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
import argparse
from dartdeepmimic import DartDeepMimicEnv
import visak_dartdeepmimic
import amc_dartdeepmimic
import os
class DartDeepMimicArgParse(argparse.ArgumentParser):
classes = {"amc": amc_dartdeepmimic.AMCDartDeepMimicEnv,
"rawqdq": visak_dartdeepmimic.VisakDartDeepMimicEnv}
def __init__(self):
super().__init__()
self.add_argument('--environment-mode', type=str, default="ddm",
help='One of "ddm" or "vdmm", specifies which env'
+ ' to instantiate')
self.add_argument('--control-skel-path', required=True,
help='Path to the control skeleton')
self.add_argument('--ref-motion-path', required=True,
help='Path to the reference motion AMC')
self.add_argument('--policy-query-frequency', required=False,
type=float, default= 30,
help="Number of times per second to query policy")
# TODO DEAD ARGUMENT
# self.add_argument('--ref-motion-dt', required=False,
# type=float, default= 1 / 120,
# help="Timestep of the motion frames")
self.add_argument('--state-mode', default=0, type=int,
help="Code for the state representation")
self.add_argument('--action-mode', type=int, required=True,
help="Code for the action representation")
self.add_argument('--visualize', default=False,
help="DOESN'T DO ANYTHING RIGHT NOW: True if you want"
+ " a window to render to")
self.add_argument('--max-torque', type=float, default=90,
help="Maximum torque")
self.add_argument('--max-angle', type=float, default=5,
help="Max magnitude of angle (in terms of pi) that "
+ "PID can output")
self.add_argument('--default-damping', type=float, default=10,
help="Default damping coefficient for joints")
self.add_argument('--default-spring', type=float, default=0,
help="Default spring stiffness for joints")
self.add_argument('--default-friction', type=float, default=20,
help="Default friction coefficient for bodies")
# TODO Dead variable, re-enable here and in dartdeepmimic
# self.add_argument('--simsteps-per-dataframe', type=int, default=10,
# help="Number of simulation steps per frame of mocap" +
# " data")
self.add_argument('--reward-cutoff', type=float, default=0.1,
help="Terminate the episode when rewards below this" +
" threshold are calculated. Should be in range (0, 1)")
self.add_argument('--window-width', type=int, default=80,
help="Window width")
self.add_argument('--window-height', type=int, default=45,
help="Window height")
self.add_argument('--pos-init-noise', type=float, default=.05,
help="Standard deviation of the position init noise")
self.add_argument('--vel-init-noise', type=float, default=.05,
help="Standart deviation of the velocity init noise")
self.add_argument('--pos-weight', type=float, default=.65,
help="Weighting for the pos difference in the reward")
self.add_argument('--pos-inner-weight', type=float, default=-2,
help="Coefficient for pos difference exponentiation in reward")
self.add_argument('--vel-weight', type=float, default=.1,
help="Weighting for the pos difference in the reward")
self.add_argument('--vel-inner-weight', type=float, default=-.1,
help="Coefficient for vel difference exponentiation in reward")
self.add_argument('--ee-weight', type=float, default=.15,
help="Weighting for the pos difference in the reward")
self.add_argument('--ee-inner-weight', type=float, default=-40,
help="Coefficient for pos difference exponentiation in reward")
self.add_argument('--com-weight', type=float, default=.1,
help="Weighting for the com difference in the reward")
self.add_argument('--com-inner-weight', type=float, default=-10,
help="Coefficient for com difference exponentiation in reward")
self.add_argument('--p-gain', type=float, default=300,
help="P for the PD controller")
self.add_argument('--d-gain', type=float, default=50,
help="D for the PD controller")
gravity_group = self.add_mutually_exclusive_group()
gravity_group.add_argument('--gravity',
dest='gravity',
action='store_true')
gravity_group.add_argument('--no-gravity',
dest='gravity',
action='store_false')
self.set_defaults(gravity=True, help="Whether to enable gravity in the world")
self_collide_group = self.add_mutually_exclusive_group()
self_collide_group.add_argument('--self-collide',
dest='selfcollide',
action='store_true')
self_collide_group.add_argument('--no-self-collide',
dest='selfcollide',
action='store_false')
self.set_defaults(self_collide=True, help="Whether to enable selfcollisions in the skeleton")
delta_group = self.add_mutually_exclusive_group()
delta_group.add_argument('--delta',
dest='delta',
action='store_true')
delta_group.add_argument('--no-delta',
dest='delta',
action='store_false')
self.set_defaults(delta=True, help="Are we in delta actions mode?")
self.add_argument('--seed', help='RNG seed', type=int,
default=None)
self.args = None
def parse_args(self):
self.args = super().parse_args()
return self.args
def get_env(self):
dir_prefix = os.path.dirname(os.path.realpath(__file__)) + "/"
return DartDeepMimicArgParse.classes[self.args.environment_mode](
skeleton_path=dir_prefix + self.args.control_skel_path,
refmotion_path=None,
statemode=1, actionmode=2,
# pos_weight=.65, pos_inner_weight=-2,
# vel_weight=.1, vel_inner_weight=-.1,
# ee_weight=.15, ee_inner_weight=-40,
# com_weight=.1, com_inner_weight=-10,
default_damping=10, default_spring=0,
default_friction=20,
visualize=self.args.visualize,
screen_width=80, screen_height=45,
# gravity=True,
self_collide=True,
delta_actions=self.args.delta,
rng_seed=self.args.seed)
return DartDeepMimicArgParse.classes[self.args.environment_mode](
skeleton_path=self.args.control_skel_path,
refmotion_path=self.args.ref_motion_path,
policy_query_frequency=self.args.policy_query_frequency,
# refmotion_dt=self.args.ref_motion_dt,
statemode=self.args.state_mode,
actionmode=self.args.action_mode,
p_gain=self.args.p_gain,
d_gain=self.args.d_gain,
pos_init_noise=self.args.pos_init_noise,
vel_init_noise=self.args.vel_init_noise,
reward_cutoff=self.args.reward_cutoff,
pos_weight=self.args.pos_weight,
pos_inner_weight=self.args.pos_inner_weight,
vel_weight=self.args.vel_weight,
vel_inner_weight=self.args.vel_inner_weight,
ee_weight=self.args.ee_weight,
ee_inner_weight=self.args.ee_inner_weight,
com_weight=self.args.com_weight,
com_inner_weight=self.args.com_inner_weight,
max_torque=self.args.max_torque,
max_angle=self.args.max_angle,
default_damping=self.args.default_damping,
default_spring=self.args.default_spring,
default_friction=self.args.default_friction,
visualize=self.args.visualize,
# simsteps_per_dataframe=self.args.simsteps_per_dataframe,
screen_width=self.args.window_width,
screen_height=self.args.window_height,
gravity=self.args.gravity,
self_collide=self.args.selfcollide,
rng_seed=self.args.seed)