Skip to content
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
286 changes: 286 additions & 0 deletions src/driverless_car_yb/main.move.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,286 @@
import pygame
import numpy as np
import cv2
import tensorflow as tf
from tensorflow.keras import layers, models
import os
from datetime import datetime

# ========================= 1. 配置参数 =========================
# 窗口设置
WINDOW_WIDTH = 800
WINDOW_HEIGHT = 600
# 无人车参数
CAR_SIZE = 40
CAR_SPEED = 5
# 环境参数
OBSTACLE_NUM = 5
OBSTACLE_SIZE = 30
# 模型参数
MODEL_PATH = "car_control_model.h5"
IMAGE_SIZE = (64, 64) # 输入模型的图像尺寸
ACTIONS = ["stop", "forward", "backward", "left", "right"] # 动作空间
ACTION_TO_IDX = {a: i for i, a in enumerate(ACTIONS)}
IDX_TO_ACTION = {i: a for i, a in enumerate(ACTIONS)}
# 数据记录(用于训练)
DATA_DIR = "training_data"
os.makedirs(DATA_DIR, exist_ok=True)

# ========================= 2. 初始化 Pygame 环境 =========================
pygame.init()
screen = pygame.display.set_mode((WINDOW_WIDTH, WINDOW_HEIGHT))
pygame.display.set_caption("深度学习无人车 - 方向键控制")
clock = pygame.time.Clock()

# 颜色定义
WHITE = (255, 255, 255)
BLUE = (0, 0, 255) # 无人车
BLACK = (0, 0, 0) # 文字/障碍物
GREEN = (0, 255, 0) # 路径
RED = (255, 0, 0) # 提示

# 字体设置
font = pygame.font.Font(None, 32)

# ========================= 3. 无人车与环境类 =========================
class Car:
def __init__(self):
# 初始位置(屏幕中心)
self.x = WINDOW_WIDTH // 2 - CAR_SIZE // 2
self.y = WINDOW_HEIGHT // 2 - CAR_SIZE // 2
self.angle = 0 # 朝向(0=向上,90=向右,180=向下,270=向左)

def draw(self, surface):
# 绘制车身(旋转的矩形)
rotated_car = pygame.transform.rotate(
pygame.Surface((CAR_SIZE, CAR_SIZE), pygame.SRCALPHA),
-self.angle # pygame 旋转方向与常规相反
)
rotated_car.fill(BLUE)
# 绘制朝向箭头
arrow_points = [
(CAR_SIZE//2, 5),
(CAR_SIZE//2 - 8, 20),
(CAR_SIZE//2 + 8, 20)
]
pygame.draw.polygon(rotated_car, BLACK, arrow_points)
# 贴图到屏幕(修正旋转后的坐标偏移)
surface.blit(
rotated_car,
(self.x - rotated_car.get_width()//2, self.y - rotated_car.get_height()//2)
)

def move(self, action):
# 根据动作更新位置和朝向
dx, dy = 0, 0
if action == "forward":
dx = CAR_SPEED * np.sin(np.radians(self.angle))
dy = -CAR_SPEED * np.cos(np.radians(self.angle))
elif action == "backward":
dx = -CAR_SPEED * np.sin(np.radians(self.angle))
dy = CAR_SPEED * np.cos(np.radians(self.angle))
elif action == "left":
self.angle = (self.angle - 5) % 360 # 左转5度
elif action == "right":
self.angle = (self.angle + 5) % 360 # 右转5度

# 边界碰撞检测
new_x = self.x + dx
new_y = self.y + dy
if 0 <= new_x <= WINDOW_WIDTH - CAR_SIZE and 0 <= new_y <= WINDOW_HEIGHT - CAR_SIZE:
self.x = new_x
self.y = new_y

class Obstacle:
def __init__(self):
# 随机生成障碍物(不与初始车位置重叠)
while True:
self.x = np.random.randint(0, WINDOW_WIDTH - OBSTACLE_SIZE)
self.y = np.random.randint(0, WINDOW_HEIGHT - OBSTACLE_SIZE)
if not (
abs(self.x - (WINDOW_WIDTH//2)) < CAR_SIZE + OBSTACLE_SIZE and
abs(self.y - (WINDOW_HEIGHT//2)) < CAR_SIZE + OBSTACLE_SIZE
):
break

def draw(self, surface):
pygame.draw.rect(surface, BLACK, (self.x, self.y, OBSTACLE_SIZE, OBSTACLE_SIZE))

# ========================= 4. 深度学习模型(CNN) =========================
def build_model():
"""构建用于无人车控制的CNN模型(输入:环境图像,输出:动作概率)"""
model = models.Sequential([
# 输入层:64x64x3(RGB图像)
layers.Conv2D(32, (3, 3), activation='relu', input_shape=(IMAGE_SIZE[0], IMAGE_SIZE[1], 3)),
layers.MaxPooling2D((2, 2)),
layers.Conv2D(64, (3, 3), activation='relu'),
layers.MaxPooling2D((2, 2)),
layers.Conv2D(128, (3, 3), activation='relu'),
layers.MaxPooling2D((2, 2)),
layers.Flatten(),
layers.Dense(128, activation='relu'),
layers.Dropout(0.5), # 防止过拟合
layers.Dense(len(ACTIONS), activation='softmax') # 输出动作概率
])
model.compile(
optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy']
)
return model

# 加载或创建模型
if os.path.exists(MODEL_PATH):
model = tf.keras.models.load_model(MODEL_PATH)
print(f"已加载现有模型:{MODEL_PATH}")
else:
model = build_model()
print("已创建新模型")

# ========================= 5. 工具函数 =========================
def get_environment_image(screen):
"""获取当前环境图像(用于模型输入)"""
# 将Pygame屏幕转换为OpenCV图像(RGB→BGR,尺寸缩放)
screen_surf = pygame.surfarray.array3d(screen)
screen_surf = cv2.cvtColor(screen_surf, cv2.COLOR_RGB2BGR)
screen_surf = cv2.resize(screen_surf, IMAGE_SIZE)
# 归一化(0-255→0-1)
return screen_surf / 255.0

def record_training_data(image, action):
"""记录训练数据(环境图像+对应的键盘动作)"""
timestamp = datetime.now().strftime("%Y%m%d%H%M%S%f")
action_idx = ACTION_TO_IDX[action]
# 保存图像
img_path = os.path.join(DATA_DIR, f"{timestamp}_{action_idx}.png")
cv2.imwrite(img_path, image * 255) # 还原为0-255保存
# 保存动作标签(单独文件,方便后续读取)
with open(os.path.join(DATA_DIR, "labels.txt"), "a") as f:
f.write(f"{img_path},{action_idx}\n")

def train_model_from_data():
"""从记录的训练数据中训练模型"""
# 读取标签文件
labels_path = os.path.join(DATA_DIR, "labels.txt")
if not os.path.exists(labels_path):
print("无训练数据,跳过训练")
return

# 加载数据和标签
X, y = [], []
with open(labels_path, "r") as f:
for line in f.readlines():
img_path, action_idx = line.strip().split(",")
if os.path.exists(img_path):
# 读取图像并预处理
img = cv2.imread(img_path)
img = cv2.resize(img, IMAGE_SIZE) / 255.0
X.append(img)
y.append(int(action_idx))

if len(X) < 10:
print("训练数据不足(至少需要10条)")
return

# 转换为numpy数组
X = np.array(X)
y = np.array(y)

# 训练模型
print(f"开始训练,样本数:{len(X)}")
model.fit(
X, y,
epochs=10,
batch_size=32,
validation_split=0.2 # 20%数据用于验证
)

# 保存模型
model.save(MODEL_PATH)
print(f"模型已保存到:{MODEL_PATH}")

# ========================= 6. 主控制逻辑 =========================
def main():
car = Car()
obstacles = [Obstacle() for _ in range(OBSTACLE_NUM)]
mode = "manual" # 初始模式:手动控制(manual)/自动控制(auto)
current_action = "stop"

running = True
while running:
# 1. 填充背景
screen.fill(WHITE)

# 2. 事件处理
for event in pygame.event.get():
if event.type == pygame.QUIT:
running = False
# 切换模式(M键)
if event.type == pygame.KEYDOWN:
if event.key == pygame.K_m:
mode = "auto" if mode == "manual" else "manual"
print(f"模式切换为:{mode}")
# 训练模型(T键)
if event.key == pygame.K_t:
print("开始训练模型...")
train_model_from_data()
# 退出(ESC键)
if event.key == pygame.K_ESCAPE:
running = False

# 3. 控制逻辑(手动/自动)
if mode == "manual":
# 手动控制:方向键映射动作
keys = pygame.key.get_pressed()
if keys[pygame.K_UP]:
current_action = "forward"
elif keys[pygame.K_DOWN]:
current_action = "backward"
elif keys[pygame.K_LEFT]:
current_action = "left"
elif keys[pygame.K_RIGHT]:
current_action = "right"
else:
current_action = "stop"

# 记录训练数据(手动模式下自动记录)
env_img = get_environment_image(screen)
record_training_data(env_img, current_action)

else:
# 自动控制:模型预测动作
env_img = get_environment_image(screen)
# 模型输入:扩展为批量维度((64,64,3)→(1,64,64,3))
env_img_batch = np.expand_dims(env_img, axis=0)
action_probs = model.predict(env_img_batch, verbose=0)[0]
current_action = IDX_TO_ACTION[np.argmax(action_probs)] # 选择概率最大的动作

# 4. 无人车移动
car.move(current_action)

# 5. 绘制元素(障碍物→无人车→提示文字)
for obs in obstacles:
obs.draw(screen)
car.draw(screen)

# 6. 显示提示信息
hint_texts = [
f"模式:{mode}(按M切换)",
f"当前动作:{current_action}",
"手动控制:↑↓←→ 移动",
"按T训练模型 | 按ESC退出"
]
for i, text in enumerate(hint_texts):
surf = font.render(text, True, BLACK if mode == "manual" else GREEN)
screen.blit(surf, (20, 20 + i*35))

# 7. 更新屏幕
pygame.display.flip()
clock.tick(30) # 30帧/秒

# 退出程序
pygame.quit()
print("程序退出")

if __name__ == "__main__":
main()
Loading