How to use efficient HRL to train ant gather

1. The entire URL of the file you are using

https://github.com/tensorflow/models/tree/master/research/efficient-hrl

2. Describe the feature you request

It seems that ant gather was not included with efficient HRL, even though maze, push and fall are. The code is very oriented to an env that requires a goal, but ant gather has no goal, I am wondering how to configure the ant gather from rllab to work with efficient HRL?

3. Additional details

I have added the gather env to the create_maze_env method like this:

if env_name == 'AntGather':
  e = AntGatherEnv(n_bins=8, sensor_span=2 * math.pi)
  e.reset()
  wrapped_env = gym_wrapper.GymWrapper(e)
  return wrapped_env

I have given it a context with meta reward defined like this: meta/Context.reward_fn = @task/plain_rewards however it does not train anything like seen in the paper. I am assuming there are some changes that need to be made since it is not a maze style env, what are those changes?

4. Are you willing to contribute it? (Yes or No)

yes