For this exercise you will need to read Chapter 12 of the NSL book. Also the supplementary reading (Metaphorical Brain 2 by M. Arbib) has a section on the topic (Section 3.4).

1. Implement the landmark learning example in NSLJ.

If you are unable to replicate the result in the NSL book try using the following guidelines.

The attractant ``tree'' is located in the origin (0, 0).

Use r=10 for the central ``tree'' and r=7 for the four other landmarks.

As a result of your work submit the following printouts:

• The NSL-program;
• The area-level graph of the matrix of weights (at the time when the system has learned the landmarks).
• Trajectory of the robot: Use an area-level plot of an auxiliary matrix in which you'll encode the current position of the system (whole matrix filled with -1, but the current position of the system, which is 1; you can use zeros to encode the position of the landmarks). For this you'll need a matrix 41x41. You can use this matrix with a little modification, to visualize the trajectories of the trained system.

2. Modify your program to try following scenario. After learning the weights with a single robot as in (1) add another robot to the environment (using the same learned weights. Note that there is no learning in this part). Introduce a repellent force that drives the robots apart. Write explicitly how the direction of movement is determined (give the equation(s)). Observe the mutual behavior of robots. Try different repellent strengths; what kind of behaviors can you get? Can the robots find the tree? Write down your observations and submit the trajectory of the robots for different type of behaviors you get. Also provide the complete NSL program.

(self study, not for credit) What if the number of landmarks is not 4, but less (3, 2)? Will the system still be able to learn the configuration of landmarks? Modify your program to make this experiment.