Tuesday, November 21, 2017

Advanced training for A.s.a. H.

Any artificially intelligent robot should be trained to feed itself, perhaps by finding a charging station and hooking up to it or perhaps by finding a brightly lit space and charging its solar batteries. A robot should also be made to discover those things that cause it pain and damage it, things like high speed collisions. Other common tasks are things like finding its way through a maze and stacking blocks. But what more complex environments, situations, and tasks should be attempted after these? Perhaps the 36 dramatic situations? (See, for example, The Thirty-six Dramatic Situations, Georges Polti, The Editor Co., 1917)

A.s.a. H. has already experienced several of the dramatic situations. The search for a recharge, whether innate or learned, would be an example of the situation  "Obtaining. Effort to obtain an object." Exploring far from the charging station might be an example of  "Daring enterprise. Adventurous expedition." The breakage of a mechanical arm while lifting would be an example of the situation "Disaster. A natural catastrophe." A multiagent system involving competition might be an example of  "Rivalry of kinsmen."

A multiagent system involving cooperation might entail "Self-sacrifice for kindred." or even "Life sacrificed to a cause." or "Deliverance. Rescue by friends."

Sunday, November 19, 2017

A.s.a.’s hierarchical memory

A.s.a.’s memory is a table of numbers. (See, for example, my blogs of 22 November 2010 and 10 February 2011.) The top three levels in the memory hierarchy were approximately:

.6 .6 .4 .4   .5 .5 .6 .4

.28 .28 .28 .28 .28 .28 .28 .28 .28 .28 .28 .28   .7 .7   .6 .6 .6   .7 .7  .6 .6 .6   1   .4 .4 .4 .4 .4  .5 .5 .5 .5   1

.7 .7   1   1   1   1   1   1   .7 .7   1   1   1   1   1   1

For one of A.s.a.’s  knowledgebases. Translating this to something like english is not easy.

Saturday, November 18, 2017

The attention problem

Although A.s.a. H. Has a large number of sensory channels as compared to most other robotic systems it still has very few inputs when compared to humans or other animals. This has probably helped with the problem of attention. But how will it scale as A.s.a. tackles more complex environments and problems?

Tuesday, November 14, 2017

Trying to know what someone else is thinking

I have tried to translate one of A.s.a. H.'s smaller concept maps into English:

sense far=(US>200), sense near=(US<75), move forward=(M1>0,M2>0), move backward=(M1<0,M2<0), turn right=(M1>0,M2<0), turn left=(M1<0,M2>0), move=(turn left), move=(turn right), move=(move backward), move=(move forward), walk=(walk forward), walk=(walk left), walk=(walk right), walk forward=(M5>0,M6>0), walk left=(M5=0, M6>0), walk right=(M5>0,M6=0), move=(walk), approach=(sense far, move forward, sense near), retreat=(sense near, move backward, sense far), decelerate=(acc<0), collision=(sense near, bump, decelerate), close hand=(M3>0), open hand=(M3<0), grasp=(proximity sense, close hand, hand force), release=(proximity sense, hand force, open hand), touch=(switch), force=(contact force), force=(hand force), force=(foot force1), force=(foot force2), force=(foot force3), force=(foot force4), push=(move, touch, contact force), arm up=(M4>0), arm down=(M4<0), force=(sense weight), lift=(arm up, sense weight), lower=(sense weight, arm down), carry=(grasp, lift, move), charge=(VB), damage=(collision, pain), health=(-damage, charge), sense=(sense far), sense=(sense near), sense=(distance), sense=(acc), sense=(touch), sense=(wind), sense=(smell), sense=(taste), sense=(temperature), sense=(hear), sense=(see), sense=(pressure), sense=(proximity sense), sense=(force), sense=(pain), sense=(charge), sense=(position), sense=(path), sense=(dust), sense=(radiation), sense=(magnetic field), sense=(direction), act=(grasp), act=(release), act=(move), act=(close hand), act=(open hand), act=(arm up), act=(arm down), act=(push), act=(lift), act=(lower), act=(carry), think=(sort file), think=(load file), think=(save file), think=(search memory), think=(case deduction), think=(case extrapolation), think=(simulation), self=(sense, think, act, health), taste=(PH, salinity), distance=(US), black=(CS~0), yellow=(CS~6), red=(CS~8.5), green(CS~4), blue=(CS~2.5), white=(CS~17), color=(black), color=(yellow), color=(red), color=(green), color=(blue), color=(white), see=(color), wind=(anemometer), temperature=(temp), hot=(temperature>80), cold=(temperature<50), hear=(sound), smell=(MQ2, MQ3, MQ4, MQ5, MQ6, MQ7, MQ8, MQ9, MQ135, O2, CO2, humidity), foot force1=(LFF), foot force2=(RFF), foot force3=(LRF), foot force4=(RRF), weight=(LFF, RFF, LRF, RRF), path=(line detection), direction=(compass), north=(compass~0), east=(compass~45), south=(compass~90), west=(compass~135), position=(lat, lon), pressure=(baro), dust=(ODS), magnetic field=(HS), radiation=(GMC)

I'm sure that this is imperfect and incomplete.

Wednesday, November 8, 2017

Concept mapping the mind of A.s.a. H.

In trying to render the meaning of A.s.a.'s hierarchical memory into English I have not always distinguished OR from AND. AND should perhaps be written as: Out1=(In1,In2) while OR would be: Out1=(In1), Out1=(In2).

After A.s.a. learned its first 100 concepts (most of the Toki Pona vocabulary) I tried to draw this up as a conventional concept map. It took me two 11" by 17" sheets of paper to complete not quite 1/3 of this.

Wednesday, November 1, 2017

A.s.a. H.'s robots, November 2017

The robots that symbol ground A.s.a. H.'s concepts are modified and changed out frequently, depending upon the experiments that are being run. Currently we are using two nearly identical mobile manipulators and a walker, all on tethers. Each mobile manipulator is a Lego eV3 pbrick mounted on wheels and with an attached robot arm. The 4 eV3 outputs command the 2 drive wheels, a servo motor that raises and lowers the arm, and a motor that opens and closes the gripper/hand. Each mobile manipulator carries 4 sensors connected to the eV3 inputs.  The walker's legs are mounted on a NXT pbrick which commands the 2 drive motors and which also carries 4 sensors connected as its inputs. A set of Vernier sensors are interfaced via LabQuests and can be carried and separately positioned using the two mobile manipulator robots.

Monday, October 30, 2017


I can currently do a good job of simulating about one third of A.s.a. H.'s sensory inputs. I have been able to do about half of the experiments with A.s.a. using simulation alone. (I am not sure how much this has been a function of the particular experiments I have happened to try.) Other tasks have involved computer simulation followed by real physical robotics experiments. I am now working on improving the fidelity of my simulations of A.s.a.'s actions (outputs).