-
Notifications
You must be signed in to change notification settings - Fork 3
/
rl_agent.py
222 lines (197 loc) · 8.73 KB
/
rl_agent.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
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
import numpy as np
from IPython import embed
import collections
import os
import tensorflow as tf
from tf_agents.agents.ddpg import critic_network
from tf_agents.agents.sac import sac_agent
from tf_agents.networks import actor_distribution_network
from tf_agents.networks import normal_projection_network
from tf_agents.networks.utils import mlp_layers
from tf_agents.policies import greedy_policy
from tf_agents.policies import py_tf_policy
from tf_agents.utils import common
from tf_agents.trajectories.time_step import TimeStep
from tensorflow.python.framework.tensor_spec import TensorSpec, BoundedTensorSpec
IMG_WIDTH = 160
IMG_HEIGHT = 90
SENSOR_DIM = 4
def normal_projection_net(action_spec,
init_action_stddev=0.35,
init_means_output_factor=0.1):
del init_action_stddev
return normal_projection_network.NormalProjectionNetwork(
action_spec,
mean_transform=None,
state_dependent_std=True,
init_means_output_factor=init_means_output_factor,
std_transform=sac_agent.std_clip_transform,
scale_distribution=True)
class SACAgent:
def __init__(
self,
root_dir,
conv_1d_layer_params=[(32, 8, 4), (64, 4, 2), (64, 3, 1)],
conv_2d_layer_params=[(32, (8, 8), 4), (64, (4, 4), 2), (64, (3, 3), 2)],
encoder_fc_layers=[256],
actor_fc_layers=[256],
critic_obs_fc_layers=[256],
critic_action_fc_layers=[256],
critic_joint_fc_layers=[256],
# Params for target update
target_update_tau=0.005,
target_update_period=1,
# Params for train
actor_learning_rate=3e-4,
critic_learning_rate=3e-4,
alpha_learning_rate=3e-4,
td_errors_loss_fn=tf.compat.v1.losses.mean_squared_error,
gamma=0.99,
reward_scale_factor=1.0,
gradient_clipping=None,
# Params for eval
eval_deterministic=False,
# Params for summaries and logging
debug_summaries=False,
summarize_grads_and_vars=False
):
'''A simple train and eval for SAC.'''
tf.compat.v1.enable_resource_variables()
root_dir = os.path.expanduser(root_dir)
policy_dir = os.path.join(root_dir, 'train', 'policy')
time_step_spec = TimeStep(
TensorSpec(shape=(), dtype=tf.int32, name='step_type'),
TensorSpec(shape=(), dtype=tf.float32, name='reward'),
BoundedTensorSpec(shape=(), dtype=tf.float32, name='discount',
minimum=np.array(0., dtype=np.float32), maximum=np.array(1., dtype=np.float32)),
collections.OrderedDict({
'sensor': BoundedTensorSpec(shape=(SENSOR_DIM,), dtype=tf.float32, name=None,
minimum=np.array(-3.4028235e+38, dtype=np.float32),
maximum=np.array(3.4028235e+38, dtype=np.float32)),
'depth': BoundedTensorSpec(shape=(IMG_HEIGHT, IMG_WIDTH, 1), dtype=tf.float32, name=None,
minimum=np.array(-1.0, dtype=np.float32),
maximum=np.array(1.0, dtype=np.float32)),
'rgb': BoundedTensorSpec(shape=(IMG_HEIGHT, IMG_WIDTH, 3), dtype=tf.float32, name=None,
minimum=np.array(-1.0, dtype=np.float32),
maximum=np.array(1.0, dtype=np.float32)),
})
)
observation_spec = time_step_spec.observation
action_spec = BoundedTensorSpec(shape=(2,), dtype=tf.float32, name=None,
minimum=np.array(-1.0, dtype=np.float32),
maximum=np.array(1.0, dtype=np.float32))
glorot_uniform_initializer = tf.compat.v1.keras.initializers.glorot_uniform()
preprocessing_layers = {}
if 'rgb' in observation_spec:
preprocessing_layers['rgb'] = tf.keras.Sequential(mlp_layers(
conv_1d_layer_params=None,
conv_2d_layer_params=conv_2d_layer_params,
fc_layer_params=encoder_fc_layers,
kernel_initializer=glorot_uniform_initializer,
))
if 'depth' in observation_spec:
preprocessing_layers['depth'] = tf.keras.Sequential(mlp_layers(
conv_1d_layer_params=None,
conv_2d_layer_params=conv_2d_layer_params,
fc_layer_params=encoder_fc_layers,
kernel_initializer=glorot_uniform_initializer,
))
if 'sensor' in observation_spec:
preprocessing_layers['sensor'] = tf.keras.Sequential(mlp_layers(
conv_1d_layer_params=None,
conv_2d_layer_params=None,
fc_layer_params=encoder_fc_layers,
kernel_initializer=glorot_uniform_initializer,
))
if len(preprocessing_layers) <= 1:
preprocessing_combiner = None
else:
preprocessing_combiner = tf.keras.layers.Concatenate(axis=-1)
actor_net = actor_distribution_network.ActorDistributionNetwork(
observation_spec,
action_spec,
preprocessing_layers=preprocessing_layers,
preprocessing_combiner=preprocessing_combiner,
fc_layer_params=actor_fc_layers,
continuous_projection_net=normal_projection_net,
kernel_initializer=glorot_uniform_initializer,
)
critic_net = critic_network.CriticNetwork(
(observation_spec, action_spec),
preprocessing_layers=preprocessing_layers,
preprocessing_combiner=preprocessing_combiner,
observation_fc_layer_params=critic_obs_fc_layers,
action_fc_layer_params=critic_action_fc_layers,
joint_fc_layer_params=critic_joint_fc_layers,
kernel_initializer=glorot_uniform_initializer,
)
global_step = tf.compat.v1.train.get_or_create_global_step()
tf_agent = sac_agent.SacAgent(
time_step_spec,
action_spec,
actor_network=actor_net,
critic_network=critic_net,
actor_optimizer=tf.compat.v1.train.AdamOptimizer(
learning_rate=actor_learning_rate),
critic_optimizer=tf.compat.v1.train.AdamOptimizer(
learning_rate=critic_learning_rate),
alpha_optimizer=tf.compat.v1.train.AdamOptimizer(
learning_rate=alpha_learning_rate),
target_update_tau=target_update_tau,
target_update_period=target_update_period,
td_errors_loss_fn=td_errors_loss_fn,
gamma=gamma,
reward_scale_factor=reward_scale_factor,
gradient_clipping=gradient_clipping,
debug_summaries=debug_summaries,
summarize_grads_and_vars=summarize_grads_and_vars,
train_step_counter=global_step)
config = tf.compat.v1.ConfigProto()
config.gpu_options.allow_growth = True
self.sess = tf.compat.v1.Session(config=config)
if eval_deterministic:
self.eval_py_policy = py_tf_policy.PyTFPolicy(greedy_policy.GreedyPolicy(tf_agent.policy))
else:
self.eval_py_policy = py_tf_policy.PyTFPolicy(tf_agent.policy)
policy_checkpointer = common.Checkpointer(
ckpt_dir=policy_dir,
policy=tf_agent.policy,
global_step=global_step)
with self.sess.as_default():
# Initialize graph.
policy_checkpointer.initialize_or_restore(self.sess)
# activate the session
obs = {
'depth': np.ones((IMG_HEIGHT, IMG_WIDTH, 1)),
'rgb': np.ones((IMG_HEIGHT, IMG_WIDTH, 3)),
'sensor': np.ones((SENSOR_DIM,))
}
action = self.act(obs)
print('activate TF session')
print('action', action)
def reset(self):
pass
def act(self, obs):
batch_obs = {}
for key in obs:
batch_obs[key] = np.expand_dims(obs[key], axis=0)
time_step = TimeStep(
np.ones(1),
np.ones(1),
np.ones(1),
batch_obs,
)
policy_state = ()
with self.sess.as_default():
action_step = self.eval_py_policy.action(time_step, policy_state)
action = action_step.action[0]
return action
if __name__ == "__main__":
obs = {
'depth': np.ones((IMG_HEIGHT, IMG_WIDTH, 1)),
'rgb': np.ones((IMG_HEIGHT, IMG_WIDTH, 3)),
'sensor': np.ones((SENSOR_DIM,))
}
agent = SACAgent(root_dir='test')
action = agent.act(obs)
print('action', action)