Robot Z
{Some things I wrote down last fall, which I discovered in transferring files to my beautiful new IBM Lenovo T-43 something-or-other; some basic psychological attributes for the robot, once we get around to it. Examples of my thinking, basically.}
{edit on March 8: I see these things, all the different behavioral routines, as being linked to sinusoidal cycles over time. In the way it's described below, 'priority' is like amplitude of these cycles, and whichever cycle has the greatest value at a given point in time will have highest priority. So, on average, a large-amplitude cycle will have highest priority, though momentary priority will depend on the various phases and frequencies of the different cycles. It's the same way we work: Eating is a large-amplitude cycle, and when you're approaching the top of the cycle you feel very hungry, and so eating is top-priority. When you've eaten, the momentary value of the routine will drop down so low that other, lesser routines with smaller amplitudes will have a chance to dominate, and you can go cycle through the blogs like you have to do every hour or so, or you can check to see what work needs to be done. Etc etc, whatever.}
In all of these basic behaviors, an important aspect must be some form of ‘inhibition of return’; once a specific behavior-event has been completed (each has some specific goal to satisfy), probably two things should happen. One, the priority of the behavior should drop, though it should not disappear. And two, the conjunction of location and behavior should somehow be inhibited; that is, the behavior should be made even less likely in that particular location. So, in conjunction with the primary floormap which is used for navigation, there could be parallel inhibitory maps which act as ‘obstacle gradients’. That is, open navigable spaces could be made more complex with, for example, low-amplitude ‘obstacle’ Gaussians placed at the location of behavior fulfillment. Over time, these Gaussians could drop in amplitude until they disappeared, leaving the area free of inhibition.
Also, some time limit or impatience should be imposed on all behaviors, and failure should also result in some, maybe lower amplitude, inhibition of return, to prevent the robot getting stuck in difficult or impossible situations.
Basic Behavior:
Rule 1: Love Green, Like Blue, Hate Red
This should be high priority with low frequency. Color-major content would be defined as some threshold proportion of the color histogram for a given tile; e.g. 25% at colormap value #25.
If green-major content is detected in one of the visual field tiles, rotate to move the content to the center column, then move forward until a) an obstacle is encountered, b) some proportion of the visual field contains green-major content, or c) the content is no longer detected. When a) or b), begin textural analysis, record position. When c), switch to next high-priority activity. {b) should be checked before a}
If blue-major content is detected in the center tile column, move forward until a), b), or c). When a) or b), carry out textural analysis and record position. {b) should be checked before a}
If red-major content is detected, move backward until a) an obstacle is encountered, b) the content is reduced to some proportion of the visual field, or c) the content is no longer detected. When a), b), or c), rotate toward the strongest summed green/blue signal and switch to next high-priority activity.
Rule 2: Like Andrew’s Shoes
This should be mid priority with mid frequency. A specific module will exist for liking Andrew’s shoes. This will consist primarily of two parts. The first part will be a pattern recognition system, trained to recognize textures, colors, etc. which correspond to the sets of shoes Andrew owns.
When the visual system thinks it has recognized an Andrew’s shoe, the second part of the module will commence. This will consist of an action phase. First, the robot will approach the object until it occupies some proportion of the visual field. Then the robot will be compelled to take some certain number of images of the object, each with some set mathematical distance from the other; that is, while maintaining the visual field proportion of the object, the robot will try to take images at different angles. Once it has reached the determined number of images, it will give some ‘satisfied’ signal (maybe a sound, or a blinking LED?), and switch to the next priority.
Rule 3: Seek out Green
This will be relatively low priority and low frequency. It should serve an important function however as a source of goal-oriented roaming. Updating the floormap will in itself be a basic behavior, but it will be of lowest priority. Seek out Green will involve creating a path to a nearby green object which has earlier been catalogued. Once this is done, minimal analysis will be done, and checked against the existing profile. Then, another nearby green target will be selected and a path created. This should continue until a higher priority behavior interferes, until all green objects are visited, or until the robot gives up due to the behavior specific time limit.
Rule 4: Orient
This would be a low frequency but high priority spiking behavior. The robot should periodically check its internal map orientation against its compass; if there’s an error, it should reorient to correct its position.
posted by Andrew H. @ 2:17 PM
4 comments
