from dm_control import suite

# Load one task:
env = suite.load(domain_name="cartpole", task_name="swingup")

# Iterate over a task set:
for domain_name, task_name in suite.BENCHMARKING:
  env = suite.load(domain_name, task_name)

# Step through an episode and print out reward, discount and observation.
action_spec = env.action_spec()
time_step = env.reset()
while not time_step.last():
  action = np.random.uniform(action_spec.minimum,
                             action_spec.maximum,
                             size=action_spec.shape)
  time_step = env.step(action)
  print(time_step.reward, time_step.discount, time_step.observation)