AI Learning to Play Tom & Jerry: Reinforcement Q-Learning

Install and import the turtle library, set up a screen with width, height, and title, and load Jerry and Tom as gif shapes for q-learning with turtles.

Learn to display the score on screen by creating a score turtle, moving it with go to, using pen up to avoid lines, and writing bold text.

Create a Jerry turtle using the turtle class, customize its shape, color, and speed, and draw a square environment with pen up and pen down.

Create an agent turtle for Tom, assign distinct names, shape it, and position it at -300, -300 to accompany Jerry and the score turtle as obstacles are introduced.

Create obstacle turtles at defined positions using a loop to form 50x50 red squares with pen up, and train Tom and Jerry with separate goals: catch and escape.

Define states and actions for a grid-like environment in a reinforcement learning setup, using q-learning to learn rewards and penalties for up, down, left, and right moves and safe states.

Explore q-tables and q-learning by training an agent to pick the best action from states and actions, using bellman updates and yaml defined prey and predator.

Define hyperparameters for reinforcement learning, including alpha learning rate, gamma discount factor, and epsilon exploration rate. Explain their roles in the Bellman update, epsilon-greedy action selection, and gradual q-table convergence.

Compare trained versus untrained Q-tables, showing direct Tom-to-Jerry movements after training and wall penalties before. Learn to save and load Q-tables with YAML and adjust epsilon.

Learn how reinforcement q-learning trains a Tom and Jerry agent pair using duration thresholds, time measurements, and looped conditions to balance catching Jerry and escaping.

Learn to use all states while training to ensure q-table convergence for every condition in a Tom & Jerry reinforcement learning setup.

Turn off rendering during training to speed up reinforcement q-learning, train with show false, iterate millions of times on states and actions, and update and clear scores while building q-tables.

Create the initial q-table for Jerry and the predator by building a condition-based dictionary with four actions (up, down, left, right). Watch training update action values toward convergence.

Apply epsilon-greedy action selection in a q-learning agent for Tom & Jerry, choosing a random action with probability epsilon and the best q-table action otherwise.

Train the turtle agent by applying up, down, left, and right actions to update the next state with 50-pixel changes and determine rewards or penalties for that state.

Define reward policy for a Tom and Jerry reinforcement learning scenario, rewarding Jerry for increasing distance from Tom and penalizing walls, obstacles, or being caught.

Learn to update a Q-table using the Bellman equation, define next states and next conditions with rewards for Jerry and Tom, and train the agent to select best actions.

Train Tom with Q-learning using predator and prey tables, epsilon-greedy actions, and Bellman updates to learn to catch Jerry while avoiding walls and obstacles; save and reuse trained data.