Maze

Train a agent to go through a basic maze, with either visual/vector observation.

This example is a more complicated version of the Grid World. The goal of the agent(yellow) is to reach the target(green) while avoiding the walls(red) within certain amount of steps.

We provide two types of training scenarios. One uses vector observation and the other uses visual observation. The results are shown below:

Maze using vector observation

The agent is able to reach the goal most of time.

Maze using visual observation

It takes longer time to train the agent with visual observation, but the result is similiar.

Go to Sourcecode

This Maze example is one of the examples in the the UnityTensorflowKeras repository. Go to the repository from the link below to install it according to the instructions.

The Maze example is located under Assets/UnityTensorflow/Examples/Maze directory.

For more information about this example, see Here.

Exercises

Exercise 1

Change the dimension of the maze and train your agent using vector observation.