AuroraMiddleware/bullet3
1
0
Fork 0
mirror of https://github.com/bulletphysics/bullet3 synced 2024-12-13 21:30:09 +00:00
Code Issues Releases Wiki Activity

add a simple DDPG agent and a policy

Browse Source
This commit is contained in:
Jie Tan 2017-04-28 15:24:53 -07:00
parent 93fc04c6c3
commit bf5915f44b
1 changed files with 6 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
examples/pybullet/gym/agents/simpleAgent.py
View File

@ -30,19 +30,17 @@ class SimpleAgent():
def _build(self):
self._agent_net = actor_net.ActorNetwork(self._actor_layer_size, self._action_size)
self._o_t = tf.placeholder(tf.float32, (31,))
self._obs = tf.placeholder(tf.float32, (31,))
with tf.name_scope('Act'):
batch_o_t = snt.nest.pack_iterable_as(
self._o_t,
snt.nest.map(
lambda x: tf.expand_dims(x, 0),
snt.nest.flatten_iterable(self._o_t)))
self._action = self._agent_net(batch_o_t)
batch_obs = snt.nest.pack_iterable_as(self._obs,
snt.nest.map(lambda x: tf.expand_dims(x, 0),
snt.nest.flatten_iterable(self._obs)))
self._action = self._agent_net(batch_obs)
saver = tf.train.Saver()
saver.restore(
sess=self._session,
save_path=self._ckpt_path)
def __call__(self, observation):
out_action = self._session.run(self._action, feed_dict={self._o_t: observation})
out_action = self._session.run(self._action, feed_dict={self._obs: observation})
return out_action[0]