Examples of using Agit training
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
agit developer 6df3bdd57a Update 'RWRL/README.md' 4 years ago
...
README.md Update 'RWRL/README.md' 4 years ago
ray_rllib_realworld_env.py Update 'RWRL/ray_rllib_realworld_env.py' 4 years ago
realworld_env.py Update 'RWRL/realworld_env.py' 4 years ago

README.md

Real-World Reinforcement Learning (RWRL) environments usage in Agit

Agit also prepacks a set of environments that can provide an evaluation of an RL algorithm’s potential applicability to real-world systems.

Currently the suite provides following environments:

| domain_name | task_name             |
|-------------|-----------------------|
| cartpole    | realworld_balance     |
| cartpole    | realworld_swingup     |
| humanoid    | realworld_stand       |
| humanoid    | realworld_walk        |
| manipulator | realworld_bring_ball  |
| manipulator | realworld_bring_peg   |
| manipulator | realworld_insert_ball |
| manipulator | realworld_insert_peg  |
| quadruped   | realworld_walk        |
| quadruped   | realworld_run         |
| walker      | realworld_stand       |
| walker      | realworld_walk        |

As mentioned above, you could access these envs as the example.

    from agit import rwrl2gym

    env = rwrl2gym.make('humanoid', 'realworld_walk',
                        combined_challenge='easy',
                        environment_kwargs={'flat_observation': True})

For specifications explanation:

https://github.com/google-research/realworldrl_suite#challenges

def make(
        domain_name,
        task_name,
        combined_challenge=None,
        safety_spec=None,
        delay_spec=None,
        noise_spec=None,
        perturb_spec=None,
        dimensionality_spec=None,
        multiobj_spec=None,
        environment_kwargs=None
):

Examples

RWRL/realworld_env.py is a simple example of realworld env.

RWRL/ray_rllib_realworld_env.py is a simple example of gym env via Ray framework.