PSY105
Aims: In the first lab class we saw that simple sensory-motor connections could produce interesting and intelligent behaviours. This lab also uses Braitenburg 'vehicles' to investigate a key property of intelligent systems - learning. During the lectures a simple classical conditioning learning mechanism has been introduced, as well as Marr’s scheme of separating the analysis of function , computation and mechanism. The lab looks at how our simple learning rule manifests in a lego robot, and how we can use synthetic models to make predictions about natural-world systems (such as ourselves).
Lab plan
At the start of the class, Tom will give a brief introduction
You will work in groups of 3 or 4, one lego robot per group, two groups per robot arena (table), and work through the tasks described in this handout.
Lab etiquette.
You will share a robot arena with a second group and you should negotiate with them to take turns to try out your robot. In general putting two robots in together will not help you to achieve the goals of the lab class! While you are waiting for your turn you can make notes, answer questions, reconfigure your robot, talk to your lab partners, or ask questions of the demonstrators. The robots are pretty robust but it is possible to break them. They are also not cheap to replace! Try to avoid dropping the robots and do not try to run them outside a robot arena, avoid getting food or liquid anywhere near the robots or the arenas.
PSY105 Autumn Semester
Completing the question sheet
As you work through the lab you should answer the questions on the sheet attached to the back of this handout. You will hand this sheet in at the end of the class and it will be reviewed and assessed as either satisfactory or unsatisfactory . If you answers are assessed as unsatisfactory you will be contacted by email and asked to attend an additional follow-up session. Your question sheets will be returned to you later in the term, however, if you are assessed as satisfactory it will not be possible to give detailed feedback. However, you can ask for help in answering the questions during the class. Note that satisfactory completion of these questions is a requirement to pass the course, but does not form part of the module assessment.
Your assessment for this lab will consist of a second set of short answer questions that you will be asked to complete in your own time and hand-in later in the term. That assignment will be marked on the 100pt scale and will count for 20% of your module mark.
PSY105: Synthetic Psychology 2 Tom Stafford, November 2010
PSY105 Autumn Semester
Task 1: Characterising untrained behaviour
1. Using the orange button on the brain brick, and the left and right scroll keys, find and select the program “Myfiles/Software Files/SuttonBarto1”.
2. Place the robot in the arena, negotiating with the other group sharing the arena so that you are not both testing robots at the same time.
3. Press the orange button to run the program. Make sure that the robot is flat on the floor when you do this.
4. Observe what the robot does - What does it respond to? How does it respond? Does it’s behaviour change over time?
Hints:
1. You do not need to rewire the robots in this lab class.
2. If your robot gets stuck in a corner, lift him out and put him back in the middle
3. What works for someone else’s robot may not work for yours.
4. If you get stuck, ask for help!
Task 2: Training a new behaviour
Now we will change the environment of the robot so there is an additional stimulus, which has a predictable relationship with the rest of the environment. Place tape around the edge of the arena, in this pattern:
Place the robot in the center. Make sure the robot is touching the floor of the arena. Select the same program “SuttonBarto1” and run the robot.
After Task 2, Tom will give a short introduction into the next tasks
PSY105: Synthetic Psychology 3 Tom Stafford, November 2010
PSY105 Autumn Semester
Task 3: Investigating the learning rate
Turn the robot on (with it touching the floor!) and then pick up the running robot
Position your robot facing a wall and mark this as the starting point
Run the robot towards the wall and, using the stop watch, record the time from starting to collision
Repeat, recording the time to collision, until the robot has learnt not to cross the line (or at least until it stops moving at the line) and hence does not collide.
Do not turn the robot off during this process, since this will wipe its memory.
Task 4: Making and testing a prediction
Using the graph from the last task as your baseline, or ‘control condition’, the final task is to test a prediction. You can use a prediction that you developed in the lecture or you can agree a new one within your group.
Use Marr’s analytical scheme to think about possible predictions
Function : What is the purpose of the behaviour? What problem does it solve? What are the inputs and outputs?
Possible predictions might focus on how we think an ideal machine for solving this problem should work and testing if our model does this.
Computation: What algorithm is used to perform the function? What calculations are made and what values are represented?
Possible predictions might focus on the elements in the learning algorithm of the model and asking how the behaviour is or should be affected if one of these elements is changed.
Mechanism: What machinery is the robot using to implement the computation?
Possible predictions might focus on how behaviour changes if we alter or remove some part.
Tom will sum-up for about ten minutes at the end of the lab-class.
Bonus Task (for those who have time)
Stopping isn’t the only response our lego robot could make to the conditioned stimulus. Select the programme “SuttonBarto2”. What does the robot do now? This new behaviour is implemented with a single simple change – what do you think this is? Hint: think back to the
Braitenberg vehicles you encountered in the first lab class. Even though the change is simple, the behaviour is far more adaptive and interesting.
Background reading for this class
Valentino Braitenberg, Vehicles: Experiments in synthetic psychology. MIT Press. 1984.
PSY105: Synthetic Psychology 4 Tom Stafford, November 2010