This project builds on the basic mbot maze runner. It is open ended for whoever takes it on. Using the same sensors as before (front, left and right) the mbot should learn, using whatever neural net or other technique, how to run the maze. Perhaps:
- the more steps been taken without finding the goal, the worse the reward function is
- inputs to the net are: how many steps so far, sensor for left, right and front
- actions are: go forward, u-turn, turn right, turn left, stop.