cartpole observation is never contained in observation_space

[vfdev]

Description:

OpenAI Gym changed the way to check if observation_space contains the observation, openai/gym#2374
and thus tasks like cartpole_discrete_balancing.py are not working anymore due to float64 vs float32 inconsistent usage:

gym-ignition/python/gym_ignition_environments/tasks/cartpole_discrete_balancing.py

Lines 66 to 68 in fa14f7f

 observation_space = gym.spaces.Box( 
     low=-obs_high, high=obs_high, dtype=np.float32 
 ) 

vs

gym-ignition/python/gym_ignition_environments/tasks/cartpole_discrete_balancing.py

Line 94 in fa14f7f

 observation = Observation(np.array([x, dx, q, dq])) 

Steps to reproduce

import gym
import numpy as np

high = np.array(
    [
        2.4,  # x
        20.0,  # dx
        np.deg2rad(12),  # q
        np.deg2rad(3 * 360),  # dq
    ]
)

# Configure the observation space
obs_high = high.copy() * 1.2
observation_space = gym.spaces.Box(
    low=-obs_high, high=obs_high, dtype=np.float32
)

x, dx, q, dq = [0.0, 0.0, 0.0, 0.0]
observation = np.array([x, dx, q, dq])

assert observation_space.contains(observation)
> AssertionError

Environment

• OS:
• GPU:
• Python:
• Version:
• Channel:
  • Stable
  • Nightly
• Installation type:
  • User
  • Developer