Skip to content

Wrappers#

The Wrapper interface is used for extending Jumanji Environment to add features like auto reset and vectorised environments. Jumanji provides wrappers to convert a Jumanji Environment to a DeepMind or Gym environment.

Jumanji to DeepMind Environment#

We can also convert our Jumanji environments to a DeepMind environment:

1
2
3
4
5
6
7
8
9
import jumanji.wrappers

env = jumanji.make("Snake-6x6-v0")
dm_env = jumanji.wrappers.JumanjiToDMEnvWrapper(env)

timestep = dm_env.reset()
action = dm_env.action_spec.generate_value()
next_timestep = dm_env.step(action)
...

Jumanji To Gym#

We can also convert our Jumanji environments to a Gym environment! Below is an example of how to convert a Jumanji environment into a Gym environment.

1
2
3
4
5
6
7
8
9
import jumanji.wrappers

env = jumanji.make("Snake-6x6-v0")
gym_env = jumanji.wrappers.JumanjiToGymWrapper(env)

obs = gym_env.reset()
action = gym_env.action_space.sample()
observation, reward, done, extra = gym_env.step(action)
...

Auto-reset an Environment#

Below is an example of how to extend the functionality of the Snake environment to automatically reset whenever the environment reaches a terminal state. The Snake game terminates when the snake hits the wall, using the AutoResetWrapper the environment will be reset once a terminal state has been reached.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
import jax.random

import jumanji.wrappers

env = jumanji.make("Snake-6x6-v0")
env = jumanji.wrappers.AutoResetWrapper(env)

key = jax.random.PRNGKey(0)
state, timestep = env.reset(key)
print("New episode")
for i in range(100):
    action = env.action_spec.generate_value()  # Returns jnp.array(0) when using Snake.
    state, timestep = env.step(state, action)
    if timestep.first():
        print("New episode")

Last update: 2024-11-01
Back to top