pythonrecess/road_crosser.py at master · AJmight/pythonrecess · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
import numpy as np
import random

# Define the states and actions
states = ['start', 'look_left', 'look_right', 'look_left_again', 'cross', 'safe', 'hit']
actions = ['look_left', 'look_right', 'cross']

# Initialize Q-table with zeros
q_table = np.zeros((len(states), len(actions)))

# Learning parameters
alpha = 0.1  # Learning rate
gamma = 0.6  # Discount factor
epsilon = 0.1  # Exploration rate

# Reward system
rewards = {
    ('start', 'look_left'): 1,
    ('look_left', 'look_right'): 1,
    ('look_right', 'look_left_again'): 1,
    ('look_left_again', 'cross'): 10,  # Big reward for correct sequence
    ('cross', 'safe'): 50,  # Successfully crossed
    ('cross', 'hit'): -100  # Got hit by car
}

# Training function
def train(num_episodes):
    for episode in range(num_episodes):
        state = 'start'
        sequence = []

        while state not in ['safe', 'hit']:
            # Epsilon-greedy action selection
            if random.uniform(0, 1) < epsilon:
                action = random.choice(actions)  # Explore
            else:
                state_idx = states.index(state)
                action_idx = np.argmax(q_table[state_idx])
                action = actions[action_idx]  # Exploit

            sequence.append((state, action))

            # Determine next state and reward
            if state == 'start' and action == 'look_left':
                next_state = 'look_left'
                reward = rewards.get((state, action), 0)
            elif state == 'look_left' and action == 'look_right':
                next_state = 'look_right'
                reward = rewards.get((state, action), 0)
            elif state == 'look_right' and action == 'look_left_again':
                next_state = 'look_left_again'
                reward = rewards.get((state, action), 0)
            elif state == 'look_left_again' and action == 'cross':
                # 90% chance of crossing safely if following the correct sequence
                if random.random() < 0.9:
                    next_state = 'safe'
                else:
                    next_state = 'hit'
                reward = rewards.get((state, action), 0)
            else:
                # Wrong action sequence - likely to get hit
                if random.random() < 0.7:
                    next_state = 'hit'
                else:
                    next_state = random.choice(states)
                reward = -10

            # Q-learning update
            state_idx = states.index(state)
            next_state_idx = states.index(next_state)
            action_idx = actions.index(action)

            old_value = q_table[state_idx, action_idx]
            next_max = np.max(q_table[next_state_idx])

            new_value = (1 - alpha) * old_value + alpha * (reward + gamma * next_max)
            q_table[state_idx, action_idx] = new_value

            state = next_state

        # Print progress
        if episode % 100 == 0:
            print(f"Episode: {episode}, Final state: {state}, Sequence: {sequence}")

# Test the trained model
def test():
    state = 'start'
    sequence = []

    print("Testing the trained model:")
    while state not in ['safe', 'hit']:
        state_idx = states.index(state)
        action_idx = np.argmax(q_table[state_idx])
        action = actions[action_idx]

        sequence.append((state, action))

        # Simulate state transition
        if state == 'start' and action == 'look_left':
            next_state = 'look_left'
        elif state == 'look_left' and action == 'look_right':
            next_state = 'look_right'
        elif state == 'look_right' and action == 'look_left_again':
            next_state = 'look_left_again'
        elif state == 'look_left_again' and action == 'cross':
            next_state = 'safe'  # Assume success in test
        else:
            next_state = 'hit'

        state = next_state

    print("Sequence executed:")
    for step in sequence:
        print(f"- {step[0]}: {step[1]}")
    print(f"Final result: {state}")

# Train the model
train(1000)

# Test the model
test()

# Show the learned Q-table
print("\nLearned Q-table:")
print(q_table)