Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

RE: Whole and Parts (and boundaries)



Title: Re: CG: Re: Re: Whole and Parts (and boundaries)

Over the last week or so I’ve found John’s comments on Whole and Parts (and boundaries) an interesting integration of some topics I knew only a smattering about and in a disconnected way.  In John’s  discussion of  natural vs. fiat boundaries he exposed some implicit  assumptions.  Others have suggested the problem of associating a boundary with just a physical boundary (e.g. a state border, a membrane or skin). Some of the time this is inappropriate, and not what we  mean. Boundaries, like part-whole relations can be seen in a much broader

context beyond a purely physical interpretation as in the state boundary by fiat. In order to go beyond the naive interpretation of a boundary as a physical object one can ask for the boundary’s function.

 

 

I’d like to add a few simple ideas to the most recent exchange with Murray Altheim on his skepticism about claims of what others think about boundaries.  The context was people’s informal discussion of boundaries using natural language.

 

John noted

>I have no real data about people talking about boundaries,
>but I have some info about people talking about walls --
>i.e., whether a wall is part of a room or a boundary
>between rooms or whether part of the wall is part of the
>room or part of the wall is part of the boundary.

>If you ask them, people have an incredible number of
>different thoughts ranging from confusion, to irritation,
>to lengthy and irrelevant discourses on the topic -- and
>essentially all of those responses are useless for any
>serious kind of application.

 

It seemed to me that one area where we do have applications that get at actual, pragmatic concepts of “boundaries” is in robotics.  Traditional robotics has explored a variety of techniques to establish perceptual and motor competence in negotiating the world.  We get to test out various ideas for perceiving bounded objects (walls etc.).  One emerging area called epigenetic robotics offers some promise of providing some more objective data about what concepts underlie boundary concepts.  Epigenetic robotics combines developmental psychology and traditional robotics to show ongoing development of behavior in robotic systems.  Epigenetic robotics takes inspiration from developmental psychology, but instead of modeling the surface behaviors of infants, it focuses on, for example, modeling the causal mechanisms and variables underlying the development of those behaviors. So rather than starting with computational approaches to say visual segmentation it utilizes psychological development as the” task domain for designing the robot and its software control architecture” . An example of this work is to design a robot that learns to improve its target-oriented reaching based on initially

having 2 things -  an accurate visual target fixation mechanism, and a reaching reflex. The robot’s reaching reflex, is based on something we see in children, an Asymmetric Tonic Neck Reflex to generate an extension of the robot’s arm, roughly in the direction of the robot’s head turn. The robot’s eye-gaze fixation control mechanism, separately tuned and not learned in this context, then develops to control the robot’s head in order to accurately fixate on an object. In this way we have a method to have the robot learn to improve its target-oriented reaching proceeded in a series of steps.

 

Here’s how it is described in a recent summary article (Core Concepts in Epigenetic Robotics) by Christopher  Prince, Nathan A. Helder, Eric J. Mislivec, & George J. Hollich

 

The goal is to “specify the overall organization for constructing the epigenetic robot, and include architectures relating to specific ontogenetic design and generic ontogenetic

design. In specific ontogenetic design, close use is made of knowledge of causal

mechanisms and variables from psychological development. In generic ontogenetic

design, the goal is to provide an ongoing emergence of behaviors in unstructured

environments with less dependence on knowledge of psychological development.”…

 

“A key, we think, lies in viewing development as a task domain. Normally, as computer scientists and engineers, when we build a system, we think of specific task requirements and construct the system to fit those requirements... For example, in constructing a computer-vision system to detect and recognize objects, we may depend on the fact that the task has non-occluded objects, perhaps with specific colors, against a neutral colored background ... In the case of an ongoing emergence of behavior, we need to reshape our thinking about design. No longer are we striving  towards task-specific design, rather we are engaging in ontogenetic design. We are designing systems that develop, and that have ongoing emerging behaviors. That is, our domain is now psychological development itself.”

 

 

 Gary Berg-Cross