如何在Open AI健身房游戏中映射自定义按键？

Question

我试图让用户使用自定义键来玩 CarRacing-v0 环境，我想我可以使用 utils.play 来实现，如下所示：

import gym
from gym.utils.play import *

play(gym.make("CarRacing-v0"))

它适用于 atari 环境，但在这种情况下我得到了

---------------------------------------------------------------------------
AssertionError                            Traceback (most recent call last)
<ipython-input-3-080385c697d2> in <module>()
      8 # play.keys_to_action = KEYWORD_TO_KEY
      9 
---> 10 play(gym.make("CarRacing-v0"))
     11 
     12 # mapping of keys

~/Documents/openai/gym/gym/utils/play.py in play(env, transpose, fps, zoom, callback, keys_to_action)
     92         else:
     93             assert False, env.spec.id + " does not have explicit key to action mapping, " + \
---> 94                           "please specify one manually"
     95     relevant_keys = set(sum(map(list, keys_to_action.keys()),[]))
     96 

AssertionError: CarRacing-v0 does not have explicit key to action mapping, please specify one manually

所以我想知道如何进行这个自定义动作映射？ 播放代码中的注释说它是一个映射keys_to_action: dict: tuple(int) -> int

atari_env 是这样做的：

KEYWORD_TO_KEY = {
            'UP':      ord('w'),
            'DOWN':    ord('s'),
            'LEFT':    ord('a'),
            'RIGHT':   ord('d'),
            'FIRE':    ord(' '),
        }

我知道 car_racing 脚本通过捕获按下 3 元素数组的按键并将该值传递给 env.step 来实现此目的。所以我在这里尝试了类似的方法：

KEYWORD_TO_KEY = {'STEER':ord('a'),'GAS':ord('w'),'BREAK':ord('s')}
play.keys_to_action = KEYWORD_TO_KEY

没成功。我知道转向是错误的，但我想我至少能让车辆向一个方向转弯。 然后我检查了将关键字重新映射到一款 Atari 游戏上的自定义组合。 游戏可以运行，但按键映射是原始的，不是我修改的。

你们知道如何正确执行此自定义键映射吗？

Answer 1

这是一个 pygame 的示例。我还在评论中添加了如何玩其他健身房环境。

import pygame
import gym
from gym.utils.play import play        
import numpy as np 
import warnings
warnings.filterwarnings('ignore')

# mapping = {(pygame.K_LEFT,): 0, (pygame.K_RIGHT,): 1}
# play(gym.make("CartPole-v0"), keys_to_action=mapping)

# mapping = {(pygame.K_LEFT,): 0, (pygame.K_RIGHT,): 2}
# play(gym.make("MountainCar-v0"), keys_to_action=mapping, noop=0)

mapping = {"w": np.array([0, 0.7, 0]),
            "a": np.array([-1, 0, 0]),
            "s": np.array([0, 0, 1]),
            "d": np.array([1, 0, 0]),
            "wa": np.array([-1, 0.7, 0]),
            "dw": np.array([1, 0.7, 0]),
            "ds": np.array([1, 0, 1]),
            "as": np.array([-1, 0, 1]),
            }
default_action = np.array([0,0,0])
play(gym.make("CarRacing-v2"), keys_to_action=mapping, noop=default_action)

Answer 2

这是一个在 play() 之外工作的示例。这使您有机会记录人类行为并将其用于学习。

def register_input():
    global quit, restart
    for event in pygame.event.get():
        if event.type == pygame.KEYDOWN:
            if event.key == pygame.K_LEFT:
                a[0] = -1.0
            if event.key == pygame.K_RIGHT:
                a[0] = +1.0
            if event.key == pygame.K_UP:
                a[1] = +1.0
            if event.key == pygame.K_DOWN:
                a[2] = +0.8  
            if event.key == pygame.K_RETURN:
                restart = True
            if event.key == pygame.K_ESCAPE:
                quit = True            
        if event.type == pygame.KEYUP:
            if event.key == pygame.K_LEFT:
                a[0] = 0
            if event.key == pygame.K_RIGHT:
                a[0] = 0
            if event.key == pygame.K_UP:
                a[1] = 0
            if event.key == pygame.K_DOWN:
                a[2] = 0            
        if event.type == pygame.QUIT:
            quit = True

env = gym.make(environment_name, render_mode="human")

quit = False
while not quit:
    env.reset()
    total_reward = 0.0
    steps = 0
    restart = False
    while True:
        register_input()
        s, r, terminated, truncated, info = env.step(action)
        total_reward += r
        if steps % 200 == 0 or terminated or truncated:
            print("\naction " + str([f"{x:+0.2f}" for x in action]))
            print(f"step {steps} total_reward {total_reward:+0.2f}")
        steps += 1
        if terminated or truncated or restart or quit:
            break
env.close()