The innovations explored in this vehicle are similar in spirit to those described for Vehicle 8, "but with a different intention this time. We will try to furnish our vehicles with a convenient set of ideas referring to the shapes of things, especially to shapes as we see them with our eyes." Basic ideas explored here are edge detection via lateral inhibition, bilateral symmetry detection, radial symmetry detection, and detection of periodicity.
Braitenberg starts with the idea that shape can be represented with "outline drawings", and that the edges of such outline shapes can be detected via lateral inhibition. (NB: This is identical in spirit to the preprocessing proposed by Marr in his discussion of early vision.) Only these detected boundaries get passed along to the next level of threshold devices, which process this represented shape in search of important, higher-order properties.
For example, Braitenberg describes a simple circuitry that can detect bilateral symmetry (right-left symmetry, in the vehicles' world). This circuitry is very similar to connectionist circuitry for similar problems. Why might bilateral symmetry be an important property to detect? Because these feature will be detected when another vehicle is directly facing (or fleeing) the observing vehicle, which might be important in gauging aggression. "The detector for bilaterally symmetrical shapes, which we have just described, proves helpful here: We may connect it to the output in such a way as to trigger the mechanisms that govern the appropriate reactions to `another vehicle facing me' or `another vehicle having me in mind'."
This raises another critical theme to Braitenberg's work: "I want you to note that something new and very important has crept into our discussion of a detector with bilateral symmetry. We decided to give our type 9 vehicles a system of connections between corresponding points on their right and left sides. In order to explain how useful such a system would be, we had to invoke not only the external appearance of other vehicles (which our vehicle might meet) but their behavior as well. Things are getting complicated: we are no longer working on individuals taken by themselves but on the members of a community in which there are complicated interactions between vehicles of the same or of different kinds."
In other words, when the vehicles attain sufficient complexity, and when there are many of them that can interact, their complexity (and the complexity of resulting interactions) in turn produces a much more complex and interesting environment for shaping vehicles. This is important both for long-term modifications via natural selection, and for short-term (individual) modifications via learning. (Question: What does this imply for the study of cognitive psychology?)
Braitenberg goes on to discuss radial symmetry detectors, which might indicate "singularities in the world." (Coupled with the detection of motion, a la Gibson, we would have reason to believe that this kind of information can be critical for guiding locomotion.) Detectors capable of picking up periodic fluctuations are also discussed (this makes me think of incorporating periodic activation functions into PDP networks).
"Taken together, vehicles of types 8 and 9 have provided much new evidence for our law of uphill analysis and downhill synthesis. A problem that taxes the minds of psychologists when they deal with real animals or humans, that of inborn concepts, found many solutions when we attacked it from the downhill, synthetic direction."
In other words, the implication of the law of downhill synthesis is that problems may be easier addressed by a synthetic approach (e.g., building simulations and watching them) instead of an analytic approach (e.g., making inferences about underlying mechanisms after watching a behaving organism).
To put this idea in the mainstream, now might be the time to do some additional reading about the advantages and disadvantages of computer simulation methods in psychology. Here's an accessible starting point:
Lewandowsky, S. (1993). The rewards and hazards of computer simulation.
Psychological science, 4, 236-243.
After reading this, you might go back to the question asked in the middle of this file, and see whether you might generate a different answer to it.