Explanaton.txt

  Issue: Asynchronous 3D rendering is not supported in the Tensorflow-based rllab.
  After a couple days of research (staring at the code) and tried out numerous methods, I think I have found
the cause:
  The "render" subroutine requires the policy object to be able to animate. However, the policy object in
the Tnesorflow-based rllab cannot be shared between multiple threads. (I couldn't figure out why unfortunately.)

  I found out that passing the entire policy object to the plotter was not necessary. All the plotter needs is the data
contained within the policy object, which in the case of the cart pole example, was the Gaussian ML Policy. 
So what I did was created two subroutines in the plotter module to work with pure data from policy. If it's the Tensorflow
training module calling the plotter, the plotter will create a policy contained within itself such that it wouldn't interfere
with the policy that's used in the training. And then pass the policy data obtained from the training loop to the policy used to 
render. In the parameterized.py module, setting the policy data requires running a Tensorflow session. But running the session that
lives in the training thread is not allowed (tf.get_default_session returns None). So I created a new session in the plotter thread. 
(I don't know why it works however.)

  The result appeared to address the issue. Both the Theano-based and Tensorflow-based rllab now support asynchronous plotting. And they now
spend about the same amount of time in one trainning session.

  The changes were made using the method of trial and error. So it might not be the most elegant way to solve the issue and it might not be
general. I'd like to submit what I have currently and have a code review. Since I have 2 finals next week. I am not sure if I will have
time to work on this anymore. But if you want me to make this more general and work on all examples, I am more than happy to do that! 

  (Currently, it only works on the trpo_cartpole.py example as I hardcoded the policy in the plotter. All the examples using Theano will still
work tho.)

  Tao