Adding experiments of gym_baselines

[zcczhang]

Hi, here are what I added/modified:

• Added GymMuJoCoSensor class in gym_sensors.py
• Added GymContinuousMuJoCoTask class in gym_task.py. However, as this is the same as GymContinuousBox2D, it can be combined into one method in the future, something like a GymContinuousTask method
• Added more env options in task_selector() and sensor_selector() in gym_tasks.py
• Added gym_baselines in project. Some notes:
  • there is a gym_model.py in gym_baselines/models. To clarify, I added this file only just for the format consistency with other baselines in the project, so it is just the same as allenact_plugins.gym_models.py so far.
  • Unlike other baselines, which use different algorithms for three envs, every gym MuJoCo env is only tested using the same model and PPO loss so far.
  • In gym Baselines, they are "1M timesteps baselines", but for now, I just use the larger timesteps to train better agents who can be visualized that they finished the tasks.
• Added gym_mujoco_tutorail in tutorials. Now there are only codes there and I will modify it to a literate experiment file with proper syntax later in case somewhere need to be changed after reviewing.
• Some other files was reformatted after I ran bash scripts/auto_format.sh

[lgtm-com[bot]]

This pull request introduces 18 alerts when merging 18a765d154c67829ec3d1dfb1a6f1493425da410 into e3ba00faeb6114777c8c0b3da2fd6a752fcc1a55 - view on LGTM.com

new alerts:

• 17 for Unused import
• 1 for Module is imported with 'import' and 'import from'

[lgtm-com[bot]]

This pull request introduces 1 alert when merging 4c083629b9529aef0c8c8635af427d4a06c45689 into e3ba00faeb6114777c8c0b3da2fd6a752fcc1a55 - view on LGTM.com

new alerts:

• 1 for Unused import

[Lucaweihs]

Nice updates! Looks good to merge to me.