DeepMind Control Suite: dm_control 1.0.13

dm_control: DeepMind Infrastructure for Physics-Based Simulation

DeepMind的软件堆栈，用于基于物理的模拟和强化学习环境，使用MuJoCo物理。

1、基准任务

for domain_name, task_name in suite.BENCHMARKING:
　　print(domain_name, task_name)
　　env = suite.load(domain_name, task_name)

　　action_spec = env.action_spec()
　　time_step = env.reset()

　　action = np.random.uniform(action_spec.minimum,
　　action_spec.maximum,
　　size=action_spec.shape)
　　time_step = env.step(action)
　　print(time_step.last(), time_step.reward, time_step.discount, time_step.observation)

• acrobot swingup
  • orientations: array(4)
    
  • velocity: array(2)
• acrobot swingup_sparse
  • orientations: array(4)
  • velocity: array(2)
• ball_in_cup catch
  • position: array(4)
  • velocity: array(4)
• cartpole balance
  • position: array(3)
  • velocity: array(2)
• cartpole balance_sparse
  • position: array(3)
  • velocity: array(2)
• cartpole swingup
  • position: array(3)
  • velocity: array(2)
• cartpole swingup_sparse
  • position: array(3)
  • velocity: array(2)
    
• cheetah run
  • position: array(8)
  • velocity: array(9)
• finger spin
  • position: array(4)
  • velocity: array(3)
  • touch: array(2)
• finger turn_easy
  • position: array(4)
  • velocity: array(3)
  • touch: array(2)
  • target_position: array(2)
  • dist_to_target: float
    
• finger turn_hard
  • position: array(4)
  • velocity: array(3)
  • touch: array(2)
  • target_position: array(2)
  • dist_to_target: float
• fish upright
  • joint_angles: array(7)
    
  • upright: float
  • velocity: array(13)
• fish swim
  • joint_angles: array(7)
    
  • upright: float
  • target: array(3)
  • velocity: array(13)
• hopper stand
  • position: array(6)
    
  • velocity: array(7)
  • touch: array(2)
• hopper hop
  • position: array(6)
    
  • velocity: array(7)
  • touch: array(2)
• humanoid stand
• humanoid walk
• humanoid run
• manipulator bring_ball
• pendulum swingup
• point_mass easy
• reacher easy
• reacher hard
• swimmer swimmer6
• swimmer swimmer15
• walker stand
• walker walk
• walker run