基于强化学习Q-Learning方法实现机器人走迷宫.zip

import numpy as np import tkinter as tk class InitLayout(tk.Tk): # gridNum格子数 gridWidth每个格子宽度 objWidth里面物体的宽度 def __init__(self, gridNum = 5, gridWidth = 80, objWidth = 50): super(InitLayout, self).__init__() self.title('机器人走迷宫') self.gridNum = gridNum self.gridWidth = gridWidth self.objWidth = objWidth self.borderSize = self.gridNum * self.gridWidth # 4种action self.action_space = ['up', 'down', 'left', 'right'] self.actions_num = len(self.action_space) # 黑洞位置, 为行列格子数的索引 self.blacks = [[0, 2], [1, 2], [2, 4], [3, 0], [3, 1], [3, 4], [4, 4], [4, 2]] # 黑洞坐标集合 self.blackCoors = [] self.start_drawing() def start_drawing(self): # 初始化 self.drawing = tk.Canvas(self, height=self.borderSize, width=self.borderSize, bg='#c9ccd0') # 画线 for col in range(0, self.borderSize, self.gridWidth): x0, y0, x1, y1 =col, 0, col, self.borderSize self.drawing.create_line(x0, y0, x1, y1) for row in range(0, self.borderSize, self.gridWidth): x0, y0, x1, y1 = 0, row, self.borderSize , row self.drawing.create_line(x0, y0, x1, y1) # 初始基点 start_pos = np.array([self.gridWidth / 2, self.gridWidth / 2]) # 循环画黑洞 for index, pos_tup in enumerate(self.blacks): col, row = pos_tup center_pos = start_pos + np.array([self.gridWidth * col, self.gridWidth * row]) hell = self.drawing.create_oval( center_pos[0] - self.objWidth / 2, center_pos[1] - self.objWidth / 2, center_pos[0] + self.objWidth / 2, center_pos[1] + self.objWidth / 2, fill='black') # 追加倒黑洞坐标中 self.blackCoors.append(self.drawing.coords(hell)) # 画终点 dist_center = start_pos + np.array([self.gridWidth * 4, self.gridWidth * 2]) self.dist = self.drawing.create_oval( dist_center[0] - self.objWidth / 2, dist_center[1] - self.objWidth / 2, dist_center[0] + self.objWidth / 2, dist_center[1] + self.objWidth / 2, fill='yellow') # 画智能体 self.rect = self.drawing.create_oval( start_pos[0] - self.objWidth / 2, start_pos[1] - self.objWidth / 2, start_pos[0] + self.objWidth / 2, start_pos[1] + self.objWidth / 2, fill='#4ed660') self.drawing.pack() def reset(self): # 重新画智能体位置 self.drawing.delete(self.rect) origin = np.array([self.gridWidth / 2, self.gridWidth / 2]) self.rect = self.drawing.create_oval( origin[0] - self.objWidth / 2, origin[1] - self.objWidth / 2, origin[0] + self.objWidth / 2, origin[1] + self.objWidth / 2, fill='#4ed660') return self.drawing.coords(self.rect) # 智能体移动 def step(self, action): s = self.drawing.coords(self.rect) agent_pos = np.array([0, 0]) # 进行移动位置 if action == 0: # up if s[1] > self.gridWidth: agent_pos[1] -= self.gridWidth elif action == 1: # down if s[1] < (self.gridNum - 1) * self.gridWidth: agent_pos[1] += self.gridWidth elif action == 2: # left if s[0] > self.gridWidth: agent_pos[0] -= self.gridWidth elif action == 3: # right if s[0] < (self.gridNum - 1) * self.gridWidth: agent_pos[0] += self.gridWidth # 移动图形 self.drawing.move(self.rect, agent_pos[0], agent_pos[1]) sig = self.drawing.coords(self.rect) # 到达终点 if sig == self.drawing.coords(self.dist): reward = 100 finished = True sig = 'finished' # 掉入黑洞 elif sig in self.blackCoors: reward = -100 finished = True sig = 'finished' else: reward = -1 finished = False return sig, reward, finished def render(self): self.update()