C++ Vector Maze.

Place an agent in 1 dimensional maze( intialize agent in a random position), the 1d maze can be of any size you can do 10 for example, have the agent go thru the maze until it finds the exit/finsihes the maze , once the agent gets to the goal it should get rewarded.

Additonally After you have that working instead of having the agent randomly make it thru the maze set up a value function V(s) = r(s) + GAMMA V(s+1),

using the value function and reward function , you can set those as additonal maze with its index holding values that get higher as you get to the goal, find a way to get out the maze optimally instead of randomly.

so first we are going to randomly have the agent go thru the maze until it finds the exit/finsihes the maze , once the agent gets to the goal it should get rewarded, then we will set it up again but this time using a value function (reinforcement learning).

How is the movement through the maze defined? Can the movement be to neighbouring locations?

Yes , movement will be through neighbouring locations.

for example look at the image i uploaded, the first index is our goal location 1, the closer you are to 1 the closer the value in that index will be, so for example if you are one box to the right or left of our goal box one, you are at 1/2 and if you are two index move away you are at 1/4 , so what you will do is compare 1/2 to 1/4 ok, 1/2 is bigger so i am going to move left instead of right since 1/2 is larger and closer to my goal exit (1).

you should be able to move both ways in the maze , so say you drop agent in index of [0,3], what you can do is say ok at this index my state value is 1/8, to my right my state value is 1/4, so i want to move right since its closer to my target goal and you repeat till you get to exit, so yes you will use neighbouring locations as movement thru maze.

Transcribed image text