Adding a New Environment

Garage uses an environment API based on the very popular OpenAI Gym interface. The main difference is that garage uses akro to describe input and output spaces, which are an extension of the gym.Space API.

If you have an OpenAI Gym compatible environment, you can wrap it in garage.envs.GymEnv to use it with garage. You can also provide its environment name id (see the example below), and let :code:`GymEnv creates the environment for you.

import gym
from garage.envs import GymEnv

my_env = GymEnv('CartPole-v1')  # shorthand for GymEnv(gym.make('CartPole-v1'))

Add an Environment Wrapper

If you would like to add an existing environment to garage, you will probably need an environment wrapper to handle environment-specific logic.

Available environment wrappers in garage are:

• garage.envs.dm_control.DmControlEnv, which supports Deepmind’s dm_control API.

• garage.envs.bullet.BulletEnv, which supports the pybullet API.

To-do list for adding a new environment wrapper:

• Specify observation_space and action_space

These state the set of valid observations and the set of valid actions. Find a detailed example in the next section.

• Implement reset and step

Allow the environment to reset and step. Find a detailed example in the next section.

• (Optional) Implement __getstate__ and __setstate__

Garage pickles the environment to save snapshots, whereas some attributes of your environment might not be pickle-able (e.g. a client-server connection). If needed, provide your own implementation to make sure your environment is pickle-able.

• (Optional) Unit Tests

If you would like to contribute your own environment back to garage, make sure to add unit tests under the directory tests/garage/envs. You can find examples of environment tests here as well.

Implement a New Environment

In the rest of this section, we will walk through an example of implementing a point robot environment using our framework. A more complete version of this environment is available as garage.envs.PointEnv.

We will implement a simple environment with 2D observations and 2D actions. The goal is to control a point robot in 2D to move it to the origin. We receive position of a point robot in the 2D plane \((x, y) \in \mathbb{R}^2\). The action is its velocity \((\dot x, \dot y) \in \mathbb{R}^2\) constrained so that \(|\dot x| \leq 0.1\) and \(|\dot y| \leq 0.1\). We encourage the robot to move to the origin by defining its reward as the negative distance to the origin: \(r(x, y) = - \sqrt{x^2 + y^2}\).

We start by creating a new file for the environment, then we declare a class inheriting from the base environment and add some imports:

import akro
import gym
import numpy as np


class PointEnv(Environment):

    # ...

For each environment, we will need to specify the set of valid observations and the set of valid actions. This is done by implementing the following property methods:

class PointEnv(Environment):

    # ...

    @property
    def observation_space(self):
        return akro.Box(low=-np.inf, high=np.inf, shape=(2,))

    @property
    def action_space(self):
        return akro.Box(low=-0.1, high=0.1, shape=(2,))

The Box space means that the observations and actions are 2D vectors with continuous values. The observations can have arbitrary values, while the actions should have magnitude at most 0.1.

Now onto the interesting part, where we actually implement the dynamics for the MDP. This is done through two methods, reset and step. The reset method randomly initializes the state of the environment according to some initial state distribution. To keep things simple, we will just sample the coordinates from a uniform distribution. The method should also return the initial observation. In our case, it will be the same as its state.

class PointEnv(Environment):

    # ...

    def reset(self):
        self._state = np.random.uniform(-1, 1, size=(2,))
        observation = np.copy(self._state)
        return observation

The step method takes an action and advances the state of the environment. It should return a Step object (which is a wrapper around namedtuple), containing the observation for the next time step, the reward, a flag indicating whether the episode is terminated after taking the step, and optional extra keyword arguments (whose values should be vectors only) for diagnostic purposes. The procedure that interfaces with the environment is responsible for calling reset after seeing that the episode is terminated.

class PointEnv(Environment):

    # ...

    def step(self, action):
        self._state = self._state + action
        x, y = self._state
        reward = - (x**2 + y**2) ** 0.5
        done = abs(x) < 0.01 and abs(y) < 0.01
        next_observation = np.copy(self._state)
        return next_observation, reward, done, None

Finally, we can implement some plotting to visualize what the MDP is doing. For simplicity, let’s just print the current state of the MDP on the terminal:

class PointEnv(Environment):

    # ...

    def render(self):
        print ('current state:', self._state)

And we’re done! We can now simulate the environment using the following diagnostic script:

python scripts/sim_env.py garage.envs.point_env --mode random

It simulates an episode of the environment with random actions, sampled from a uniform distribution within the defined action bounds.

You could also train a neural network policy to solve the task, which is probably overkill. To do so, create a new script with the following content (we will use stub mode):

from garage import wrap_experiment
from garage.envs import PointEnv
from garage.envs import normalize
from garage.experiment.deterministic import set_seed
from garage.np.baselines import LinearFeatureBaseline
from garage.tf.algos import TRPO
from garage.tf.policies import CategoricalMLPPolicy
from garage.trainer import TFTrainer


@wrap_experiment
def trpo_point(ctxt=None, seed=1):
    set_seed(seed)
    with TFTrainer(ctxt) as trainer:
        env = normalize(PointEnv())

        policy = CategoricalMLPPolicy(name='policy',
                                      env_spec=env.spec,
                                      hidden_sizes=(32, 32))

        baseline = LinearFeatureBaseline(env_spec=env.spec)

        algo = TRPO(env_spec=env.spec,
                    policy=policy,
                    baseline=baseline,
                    discount=0.99,
                    max_kl_step=0.01)

        trainer.setup(algo, env)
        trainer.train(n_epochs=100, batch_size=4000)


trpo_point()

Assume that the file is examples/tf/trpo_point.py. You can then run the script:

python examples/tf/trpo_point.py

---

This page was authored by K.R. Zentner (@krzentner), with contributions from Ryan Julian (@ryanjulian), Jonathon Shen (@jonashen), Rocky Duan (@dementrock), Eric Yihan Chen (@AiRuiChen).