Revenge of the SOM

• Apply Q-Learning to automatically learn the four behaviors below. Instead of discretizing the state space, however, use a Self-Organizing Map to encode the sensor inputs into a state. You should use the same sensor configurations, action sets, and reward functions you used in the previous assignment. Select one task for which you will experiment with three different sizes of SOM. Use the most satisfactory of the sizes for the remaining experiments.
  • Object avoidance
  • Wandering (with directed motion)
  • Wall-following
  • Pursuit
  • Light-finding
• Write a report detailing your results. Discuss whether Q-Learning with or without a SOM is preferable. Also assess the degree to which machine learning is even appropriate for these tasks, relative to a hand-coded implementation. Support your claims with data and observations from your experiments.