This is a simple implementation of Q-learning algorithm to solve the wolf-goat-cabbage problem. The problem is a well-known puzzle in artificial intelligence, and it is used to illustrate the power of state-space search algorithms. The problem can be stated as follows: A farmer with a wolf, a goat, and a cabbage come to the edge of a river they wish to cross. There is a boat at the river's edge that only the farmer can row. The farmer can take at most one other object besides himself on a crossing, but if the wolf is ever left with the goat, the wolf will eat the goat; similarly, if the goat is left with the cabbage, the goat will eat the cabbage. Devise a sequence of crossings of the river so that all four characters arrive safely on the other side of the river.

• Python 3.6
• gym
• numpy

pip3 install gym
pip3 install numpy

python3 main.py

The Q-learning algorithm is a model-free, online, off-policy reinforcement learning method. The main idea of Q-learning is to use the Bellman equation as an iterative update rule for action-value pairs. The algorithm in Wolf-Goat-Cabbage problem is as follows:

• Initialize Q(s,a) arbitrarily
• Repeat (for each episode):
  • Initialize s
  • Repeat (for each step of episode):
    • Choose a from s using policy derived from Q (e.g., epsilon-greedy)
    • Take action a, observe r, s'
    • Q(s,a) <- Q(s,a) + alpha[r + gamma * max(Q(s',a')) - Q(s,a)]
    • s <- s'
  • until s is terminal