訓練一個玩馬力歐的強化學習智慧體

建立日期：2020 年 12 月 17 日 | 最後更新：2024 年 2 月 5 日 | 最後驗證：未驗證

作者： Yuansong Feng, Suraj Subramanian, Howard Wang, Steven Guo.

本教程將引導你瞭解深度強化學習的基礎知識。學完本教程，你將能夠實現一個由 AI 驅動的馬力歐（使用雙深度 Q 網路），它能夠自己玩遊戲。

雖然本教程不需要強化學習的先驗知識，但你可以透過這些強化學習概念來熟悉自己，並把這份方便的備忘單作為你的參考。完整程式碼可在此處檢視。

%%bash
pip install gym-super-mario-bros==7.4.0
pip install tensordict==0.3.0
pip install torchrl==0.3.0

import torch
from torch import nn
from torchvision import transforms as T
from PIL import Image
import numpy as np
from pathlib import Path
from collections import deque
import random, datetime, os

# Gym is an OpenAI toolkit for RL
import gym
from gym.spaces import Box
from gym.wrappers import FrameStack

# NES Emulator for OpenAI Gym
from nes_py.wrappers import JoypadSpace

# Super Mario environment for OpenAI Gym
import gym_super_mario_bros

from tensordict import TensorDict
from torchrl.data import TensorDictReplayBuffer, LazyMemmapStorage

強化學習定義

環境 智慧體與之互動並從中學習的世界。

動作 \(a\) : 智慧體對環境做出響應的方式。所有可能動作的集合稱為動作空間。

狀態 \(s\) : 環境當前的特徵。環境可能處於的所有狀態的集合稱為狀態空間。

獎勵 \(r\) : 獎勵是環境對智慧體的關鍵反饋。它是驅動智慧體學習和改變其未來動作的動力。跨多個時間步長的獎勵總和稱為回報。

最優動作價值函式 \(Q^*(s,a)\) : 給出了從狀態 \(s\) 開始，採取任意動作 \(a\)，然後在未來每個時間步長都採取使回報最大化的動作時，所期望的回報。\(Q\) 可以說代表了動作在某個狀態下的“質量”。我們嘗試近似這個函式。

環境

初始化環境

在馬力歐遊戲中，環境由水管、蘑菇和其他組成部分構成。

當馬力歐採取動作時，環境會返回改變後的（下一個）狀態、獎勵和其他資訊作為響應。

# Initialize Super Mario environment (in v0.26 change render mode to 'human' to see results on the screen)
if gym.__version__ < '0.26':
    env = gym_super_mario_bros.make("SuperMarioBros-1-1-v0", new_step_api=True)
else:
    env = gym_super_mario_bros.make("SuperMarioBros-1-1-v0", render_mode='rgb', apply_api_compatibility=True)

# Limit the action-space to
#   0. walk right
#   1. jump right
env = JoypadSpace(env, [["right"], ["right", "A"]])

env.reset()
next_state, reward, done, trunc, info = env.step(action=0)
print(f"{next_state.shape},\n {reward},\n {done},\n {info}")

預處理環境

環境資料在 next_state 中返回給智慧體。如上所示，每個狀態由一個大小為 [3, 240, 256] 的陣列表示。通常，這比我們的智慧體所需的資訊更多；例如，馬力歐的動作並不取決於水管或天空的顏色！

我們使用包裝器 (Wrappers) 在將環境資料傳送給智慧體之前對其進行預處理。

GrayScaleObservation 是一種常見的包裝器，用於將 RGB 影像轉換為灰度影像；這樣做可以在不丟失有用資訊的情況下減小狀態表示的大小。現在每個狀態的大小為：[1, 240, 256]

ResizeObservation 將每個觀測結果下采樣為一個方形影像。新的大小為：[1, 84, 84]

SkipFrame 是一個繼承自 gym.Wrapper 並實現了 step() 函式的自定義包裝器。由於連續幀變化不大，我們可以跳過 n 箇中間幀而不會丟失太多資訊。第 n 幀累積了跳過的每一幀中獲得的獎勵。

FrameStack 是一個包裝器，它允許我們將環境中連續的幀壓縮到單個觀測點，以饋送給我們的學習模型。透過這種方式，我們可以根據馬力歐在前幾個幀中的移動方向來判斷他是落地還是跳躍。

class SkipFrame(gym.Wrapper):
    def __init__(self, env, skip):
        """Return only every `skip`-th frame"""
        super().__init__(env)
        self._skip = skip

    def step(self, action):
        """Repeat action, and sum reward"""
        total_reward = 0.0
        for i in range(self._skip):
            # Accumulate reward and repeat the same action
            obs, reward, done, trunk, info = self.env.step(action)
            total_reward += reward
            if done:
                break
        return obs, total_reward, done, trunk, info


class GrayScaleObservation(gym.ObservationWrapper):
    def __init__(self, env):
        super().__init__(env)
        obs_shape = self.observation_space.shape[:2]
        self.observation_space = Box(low=0, high=255, shape=obs_shape, dtype=np.uint8)

    def permute_orientation(self, observation):
        # permute [H, W, C] array to [C, H, W] tensor
        observation = np.transpose(observation, (2, 0, 1))
        observation = torch.tensor(observation.copy(), dtype=torch.float)
        return observation

    def observation(self, observation):
        observation = self.permute_orientation(observation)
        transform = T.Grayscale()
        observation = transform(observation)
        return observation


class ResizeObservation(gym.ObservationWrapper):
    def __init__(self, env, shape):
        super().__init__(env)
        if isinstance(shape, int):
            self.shape = (shape, shape)
        else:
            self.shape = tuple(shape)

        obs_shape = self.shape + self.observation_space.shape[2:]
        self.observation_space = Box(low=0, high=255, shape=obs_shape, dtype=np.uint8)

    def observation(self, observation):
        transforms = T.Compose(
            [T.Resize(self.shape, antialias=True), T.Normalize(0, 255)]
        )
        observation = transforms(observation).squeeze(0)
        return observation


# Apply Wrappers to environment
env = SkipFrame(env, skip=4)
env = GrayScaleObservation(env)
env = ResizeObservation(env, shape=84)
if gym.__version__ < '0.26':
    env = FrameStack(env, num_stack=4, new_step_api=True)
else:
    env = FrameStack(env, num_stack=4)

將上述包裝器應用於環境後，最終的包裝狀態由 4 個堆疊在一起的灰度連續幀組成，如左側影像所示。每次馬力歐採取動作時，環境都會返回一個具有這種結構的狀態作為響應。該結構由一個大小為 [4, 84, 84] 的 3-D 陣列表示。

智慧體

我們建立一個名為 Mario 的類來表示遊戲中的智慧體。馬力歐應該能夠做到

• 根據當前（環境）狀態，按照最優動作策略行動。

• 記住經驗。經驗 = (當前狀態, 當前動作, 獎勵, 下一個狀態)。馬力歐會快取並稍後回憶他的經驗，以更新他的動作策略。

• 隨著時間推移學習更好的動作策略

class Mario:
    def __init__():
        pass

    def act(self, state):
        """Given a state, choose an epsilon-greedy action"""
        pass

    def cache(self, experience):
        """Add the experience to memory"""
        pass

    def recall(self):
        """Sample experiences from memory"""
        pass

    def learn(self):
        """Update online action value (Q) function with a batch of experiences"""
        pass

在接下來的部分中，我們將填充馬力歐的引數並定義他的函式。

行動

對於任何給定的狀態，智慧體可以選擇執行最優動作（利用）或隨機動作（探索）。

馬力歐以 self.exploration_rate 的機率隨機探索；當他選擇利用時，他依賴 MarioNet（在 學習 部分實現）來提供最優動作。

class Mario:
    def __init__(self, state_dim, action_dim, save_dir):
        self.state_dim = state_dim
        self.action_dim = action_dim
        self.save_dir = save_dir

        self.device = "cuda" if torch.cuda.is_available() else "cpu"

        # Mario's DNN to predict the most optimal action - we implement this in the Learn section
        self.net = MarioNet(self.state_dim, self.action_dim).float()
        self.net = self.net.to(device=self.device)

        self.exploration_rate = 1
        self.exploration_rate_decay = 0.99999975
        self.exploration_rate_min = 0.1
        self.curr_step = 0

        self.save_every = 5e5  # no. of experiences between saving Mario Net

    def act(self, state):
        """
    Given a state, choose an epsilon-greedy action and update value of step.

    Inputs:
    state(``LazyFrame``): A single observation of the current state, dimension is (state_dim)
    Outputs:
    ``action_idx`` (``int``): An integer representing which action Mario will perform
    """
        # EXPLORE
        if np.random.rand() < self.exploration_rate:
            action_idx = np.random.randint(self.action_dim)

        # EXPLOIT
        else:
            state = state[0].__array__() if isinstance(state, tuple) else state.__array__()
            state = torch.tensor(state, device=self.device).unsqueeze(0)
            action_values = self.net(state, model="online")
            action_idx = torch.argmax(action_values, axis=1).item()

        # decrease exploration_rate
        self.exploration_rate *= self.exploration_rate_decay
        self.exploration_rate = max(self.exploration_rate_min, self.exploration_rate)

        # increment step
        self.curr_step += 1
        return action_idx

快取和回憶

這兩個函式作為馬力歐的“記憶”過程。

cache()：每次馬力歐執行動作時，他都會將 經驗 儲存到他的記憶中。他的經驗包括當前的狀態、執行的動作、從動作獲得的獎勵、下一個狀態以及遊戲是否結束。

recall()：馬力歐從他的記憶中隨機抽取一批經驗，並使用這些經驗來學習遊戲。

class Mario(Mario):  # subclassing for continuity
    def __init__(self, state_dim, action_dim, save_dir):
        super().__init__(state_dim, action_dim, save_dir)
        self.memory = TensorDictReplayBuffer(storage=LazyMemmapStorage(100000, device=torch.device("cpu")))
        self.batch_size = 32

    def cache(self, state, next_state, action, reward, done):
        """
        Store the experience to self.memory (replay buffer)

        Inputs:
        state (``LazyFrame``),
        next_state (``LazyFrame``),
        action (``int``),
        reward (``float``),
        done(``bool``))
        """
        def first_if_tuple(x):
            return x[0] if isinstance(x, tuple) else x
        state = first_if_tuple(state).__array__()
        next_state = first_if_tuple(next_state).__array__()

        state = torch.tensor(state)
        next_state = torch.tensor(next_state)
        action = torch.tensor([action])
        reward = torch.tensor([reward])
        done = torch.tensor([done])

        # self.memory.append((state, next_state, action, reward, done,))
        self.memory.add(TensorDict({"state": state, "next_state": next_state, "action": action, "reward": reward, "done": done}, batch_size=[]))

    def recall(self):
        """
        Retrieve a batch of experiences from memory
        """
        batch = self.memory.sample(self.batch_size).to(self.device)
        state, next_state, action, reward, done = (batch.get(key) for key in ("state", "next_state", "action", "reward", "done"))
        return state, next_state, action.squeeze(), reward.squeeze(), done.squeeze()

學習

馬力歐內部使用DDQN 演算法。DDQN 使用兩個卷積神經網路 (ConvNets) - \(Q_{online}\) 和 \(Q_{target}\) - 它們獨立地近似最優動作價值函式。

在我們的實現中，我們在 \(Q_{online}\) 和 \(Q_{target}\) 之間共享特徵生成器 features，但為它們各自維護獨立的 FC 分類器。\(\theta_{target}\)（\(Q_{target}\) 的引數）被凍結，以防止透過反向傳播進行更新。相反，它會定期與 \(\theta_{online}\) 同步（稍後詳細介紹）。

神經網路

class MarioNet(nn.Module):
    """mini CNN structure
  input -> (conv2d + relu) x 3 -> flatten -> (dense + relu) x 2 -> output
  """

    def __init__(self, input_dim, output_dim):
        super().__init__()
        c, h, w = input_dim

        if h != 84:
            raise ValueError(f"Expecting input height: 84, got: {h}")
        if w != 84:
            raise ValueError(f"Expecting input width: 84, got: {w}")

        self.online = self.__build_cnn(c, output_dim)

        self.target = self.__build_cnn(c, output_dim)
        self.target.load_state_dict(self.online.state_dict())

        # Q_target parameters are frozen.
        for p in self.target.parameters():
            p.requires_grad = False

    def forward(self, input, model):
        if model == "online":
            return self.online(input)
        elif model == "target":
            return self.target(input)

    def __build_cnn(self, c, output_dim):
        return nn.Sequential(
            nn.Conv2d(in_channels=c, out_channels=32, kernel_size=8, stride=4),
            nn.ReLU(),
            nn.Conv2d(in_channels=32, out_channels=64, kernel_size=4, stride=2),
            nn.ReLU(),
            nn.Conv2d(in_channels=64, out_channels=64, kernel_size=3, stride=1),
            nn.ReLU(),
            nn.Flatten(),
            nn.Linear(3136, 512),
            nn.ReLU(),
            nn.Linear(512, output_dim),
        )

TD 估計與 TD 目標

學習過程中涉及兩個值

TD 估計 - 給定狀態 \(s\) 的預測最優 \(Q^*\)

\[{TD}_e = Q_{online}^*(s,a)\]

TD 目標 - 當前獎勵和下一個狀態 \(s'\) 中的估計 \(Q^*\) 的聚合

\[a' = argmax_{a} Q_{online}(s', a)\]

\[{TD}_t = r + \gamma Q_{target}^*(s',a')\]

因為我們不知道下一個動作 \(a'\) 是什麼，所以我們使用在下一個狀態 \(s'\) 中使 \(Q_{online}\) 最大化的動作 \(a'\)。

注意，我們在 td_target() 上使用了 @torch.no_grad() 裝飾器，以在此處停用梯度計算（因為我們不需要對 \(\theta_{target}\) 進行反向傳播）。

class Mario(Mario):
    def __init__(self, state_dim, action_dim, save_dir):
        super().__init__(state_dim, action_dim, save_dir)
        self.gamma = 0.9

    def td_estimate(self, state, action):
        current_Q = self.net(state, model="online")[
            np.arange(0, self.batch_size), action
        ]  # Q_online(s,a)
        return current_Q

    @torch.no_grad()
    def td_target(self, reward, next_state, done):
        next_state_Q = self.net(next_state, model="online")
        best_action = torch.argmax(next_state_Q, axis=1)
        next_Q = self.net(next_state, model="target")[
            np.arange(0, self.batch_size), best_action
        ]
        return (reward + (1 - done.float()) * self.gamma * next_Q).float()

更新模型

當馬力歐從他的重放緩衝區中取樣輸入時，我們計算 \(TD_t\) 和 \(TD_e\)，並透過反向傳播這個損失來更新 \(Q_{online}\) 的引數 \(\theta_{online}\)（\(\alpha\) 是傳遞給 optimizer 的學習率 lr）

\[\theta_{online} \leftarrow \theta_{online} + \alpha \nabla(TD_e - TD_t)\]

\(\theta_{target}\) 不透過反向傳播更新。相反，我們定期將 \(\theta_{online}\) 複製到 \(\theta_{target}\)

\[\theta_{target} \leftarrow \theta_{online}\]

class Mario(Mario):
    def __init__(self, state_dim, action_dim, save_dir):
        super().__init__(state_dim, action_dim, save_dir)
        self.optimizer = torch.optim.Adam(self.net.parameters(), lr=0.00025)
        self.loss_fn = torch.nn.SmoothL1Loss()

    def update_Q_online(self, td_estimate, td_target):
        loss = self.loss_fn(td_estimate, td_target)
        self.optimizer.zero_grad()
        loss.backward()
        self.optimizer.step()
        return loss.item()

    def sync_Q_target(self):
        self.net.target.load_state_dict(self.net.online.state_dict())

儲存檢查點

class Mario(Mario):
    def save(self):
        save_path = (
            self.save_dir / f"mario_net_{int(self.curr_step // self.save_every)}.chkpt"
        )
        torch.save(
            dict(model=self.net.state_dict(), exploration_rate=self.exploration_rate),
            save_path,
        )
        print(f"MarioNet saved to {save_path} at step {self.curr_step}")

整合所有部分

class Mario(Mario):
    def __init__(self, state_dim, action_dim, save_dir):
        super().__init__(state_dim, action_dim, save_dir)
        self.burnin = 1e4  # min. experiences before training
        self.learn_every = 3  # no. of experiences between updates to Q_online
        self.sync_every = 1e4  # no. of experiences between Q_target & Q_online sync

    def learn(self):
        if self.curr_step % self.sync_every == 0:
            self.sync_Q_target()

        if self.curr_step % self.save_every == 0:
            self.save()

        if self.curr_step < self.burnin:
            return None, None

        if self.curr_step % self.learn_every != 0:
            return None, None

        # Sample from memory
        state, next_state, action, reward, done = self.recall()

        # Get TD Estimate
        td_est = self.td_estimate(state, action)

        # Get TD Target
        td_tgt = self.td_target(reward, next_state, done)

        # Backpropagate loss through Q_online
        loss = self.update_Q_online(td_est, td_tgt)

        return (td_est.mean().item(), loss)

日誌記錄

import numpy as np
import time, datetime
import matplotlib.pyplot as plt


class MetricLogger:
    def __init__(self, save_dir):
        self.save_log = save_dir / "log"
        with open(self.save_log, "w") as f:
            f.write(
                f"{'Episode':>8}{'Step':>8}{'Epsilon':>10}{'MeanReward':>15}"
                f"{'MeanLength':>15}{'MeanLoss':>15}{'MeanQValue':>15}"
                f"{'TimeDelta':>15}{'Time':>20}\n"
            )
        self.ep_rewards_plot = save_dir / "reward_plot.jpg"
        self.ep_lengths_plot = save_dir / "length_plot.jpg"
        self.ep_avg_losses_plot = save_dir / "loss_plot.jpg"
        self.ep_avg_qs_plot = save_dir / "q_plot.jpg"

        # History metrics
        self.ep_rewards = []
        self.ep_lengths = []
        self.ep_avg_losses = []
        self.ep_avg_qs = []

        # Moving averages, added for every call to record()
        self.moving_avg_ep_rewards = []
        self.moving_avg_ep_lengths = []
        self.moving_avg_ep_avg_losses = []
        self.moving_avg_ep_avg_qs = []

        # Current episode metric
        self.init_episode()

        # Timing
        self.record_time = time.time()

    def log_step(self, reward, loss, q):
        self.curr_ep_reward += reward
        self.curr_ep_length += 1
        if loss:
            self.curr_ep_loss += loss
            self.curr_ep_q += q
            self.curr_ep_loss_length += 1

    def log_episode(self):
        "Mark end of episode"
        self.ep_rewards.append(self.curr_ep_reward)
        self.ep_lengths.append(self.curr_ep_length)
        if self.curr_ep_loss_length == 0:
            ep_avg_loss = 0
            ep_avg_q = 0
        else:
            ep_avg_loss = np.round(self.curr_ep_loss / self.curr_ep_loss_length, 5)
            ep_avg_q = np.round(self.curr_ep_q / self.curr_ep_loss_length, 5)
        self.ep_avg_losses.append(ep_avg_loss)
        self.ep_avg_qs.append(ep_avg_q)

        self.init_episode()

    def init_episode(self):
        self.curr_ep_reward = 0.0
        self.curr_ep_length = 0
        self.curr_ep_loss = 0.0
        self.curr_ep_q = 0.0
        self.curr_ep_loss_length = 0

    def record(self, episode, epsilon, step):
        mean_ep_reward = np.round(np.mean(self.ep_rewards[-100:]), 3)
        mean_ep_length = np.round(np.mean(self.ep_lengths[-100:]), 3)
        mean_ep_loss = np.round(np.mean(self.ep_avg_losses[-100:]), 3)
        mean_ep_q = np.round(np.mean(self.ep_avg_qs[-100:]), 3)
        self.moving_avg_ep_rewards.append(mean_ep_reward)
        self.moving_avg_ep_lengths.append(mean_ep_length)
        self.moving_avg_ep_avg_losses.append(mean_ep_loss)
        self.moving_avg_ep_avg_qs.append(mean_ep_q)

        last_record_time = self.record_time
        self.record_time = time.time()
        time_since_last_record = np.round(self.record_time - last_record_time, 3)

        print(
            f"Episode {episode} - "
            f"Step {step} - "
            f"Epsilon {epsilon} - "
            f"Mean Reward {mean_ep_reward} - "
            f"Mean Length {mean_ep_length} - "
            f"Mean Loss {mean_ep_loss} - "
            f"Mean Q Value {mean_ep_q} - "
            f"Time Delta {time_since_last_record} - "
            f"Time {datetime.datetime.now().strftime('%Y-%m-%dT%H:%M:%S')}"
        )

        with open(self.save_log, "a") as f:
            f.write(
                f"{episode:8d}{step:8d}{epsilon:10.3f}"
                f"{mean_ep_reward:15.3f}{mean_ep_length:15.3f}{mean_ep_loss:15.3f}{mean_ep_q:15.3f}"
                f"{time_since_last_record:15.3f}"
                f"{datetime.datetime.now().strftime('%Y-%m-%dT%H:%M:%S'):>20}\n"
            )

        for metric in ["ep_lengths", "ep_avg_losses", "ep_avg_qs", "ep_rewards"]:
            plt.clf()
            plt.plot(getattr(self, f"moving_avg_{metric}"), label=f"moving_avg_{metric}")
            plt.legend()
            plt.savefig(getattr(self, f"{metric}_plot"))

開始玩吧！

在此示例中，我們執行訓練迴圈 40 個回合，但要讓馬力歐真正學會他所在世界的玩法，我們建議執行迴圈至少 40,000 個回合！

use_cuda = torch.cuda.is_available()
print(f"Using CUDA: {use_cuda}")
print()

save_dir = Path("checkpoints") / datetime.datetime.now().strftime("%Y-%m-%dT%H-%M-%S")
save_dir.mkdir(parents=True)

mario = Mario(state_dim=(4, 84, 84), action_dim=env.action_space.n, save_dir=save_dir)

logger = MetricLogger(save_dir)

episodes = 40
for e in range(episodes):

    state = env.reset()

    # Play the game!
    while True:

        # Run agent on the state
        action = mario.act(state)

        # Agent performs action
        next_state, reward, done, trunc, info = env.step(action)

        # Remember
        mario.cache(state, next_state, action, reward, done)

        # Learn
        q, loss = mario.learn()

        # Logging
        logger.log_step(reward, loss, q)

        # Update state
        state = next_state

        # Check if end of game
        if done or info["flag_get"]:
            break

    logger.log_episode()

    if (e % 20 == 0) or (e == episodes - 1):
        logger.record(episode=e, epsilon=mario.exploration_rate, step=mario.curr_step)

結論

在本教程中，我們展示瞭如何使用 PyTorch 訓練一個玩遊戲的 AI。你可以使用相同的方法來訓練 AI 玩 OpenAI gym 中的任何遊戲。希望你喜歡本教程，歡迎隨時透過我們的 github 與我們聯絡！