malak123A456/q-learning-gridworld

1import numpy as np
2import random
3import matplotlib.pyplot as plt
4
5# Parameters
6alpha = 0.1  # Learning rate
7gamma = 0.9  # Discount factor
8epsilon = 1.0  # Exploration rate
9epsilon_decay = 0.995
10min_epsilon = 0.01
11episodes = 1000
12grid_size = 5
13
14# Initialize Q-table
15q_table = np.zeros((grid_size, grid_size, 4))  # 4 actions: up, down, left, right
16
17# Environment definition
18start_state = (0, 0)
19goal_state = (4, 4)
20obstacles = [(2, 2), (3, 3)]
21
22# Rewards
23def get_reward(state):
24    if state == goal_state:
25        return 100
26    elif state in obstacles:
27        return -10
28    return -1
29
30# Get next state
31def get_next_state(state, action):
32    x, y = state
33    if action == 0 and x > 0:  # Up
34        x -= 1
35    elif action == 1 and x < grid_size - 1:  # Down
36        x += 1
37    elif action == 2 and y > 0:  # Left
38        y -= 1
39    elif action == 3 and y < grid_size - 1:  # Right
40        y += 1
41    return (x, y)
42
43# Train Q-learning
44for episode in range(episodes):
45    state = start_state
46    done = False
47    while not done:
48        # Choose action using epsilon-greedy
49        if random.uniform(0, 1) < epsilon:
50            action = random.randint(0, 3)  # Explore
51        else:
52            action = np.argmax(q_table[state])  # Exploit
53
54        # Take action
55        next_state = get_next_state(state, action)
56        reward = get_reward(next_state)
57
58        # Update Q-value
59        best_next_action = np.argmax(q_table[next_state])
60        q_table[state][action] += alpha * (
61            reward + gamma * q_table[next_state][best_next_action] - q_table[state][action]
62        )
63
64        state = next_state
65
66        # Check if episode is done
67        if state == goal_state or state in obstacles:
68            done = True
69
70    # Decay epsilon
71    epsilon = max(min_epsilon, epsilon * epsilon_decay)
72
73# Visualize optimal policy
74policy = np.zeros((grid_size, grid_size), dtype=str)
75directions = ['↑', '↓', '←', '→']
76for i in range(grid_size):
77    for j in range(grid_size):
78        policy[i, j] = directions[np.argmax(q_table[(i, j)])]
79policy[goal_state] = 'G'
80for obs in obstacles:
81    policy[obs] = 'X'
82policy[start_state] = 'S'
83
84print("Optimal Policy:")
85print(policy)
86