1. robot ai
- Posted by John McAdam <johnmcadam at CLIX.PT> Jan 30, 2001
- 484 views
The robots can learn from their experience. If they vanquished a foe in one situation and then are in a similar situation but with a different foe - they can use pattern recognition tecniques to guess that a similar response should be tried. If that doesn't work they can learn to recognize the differences. You would probably need a shoot-em- up type of game so that the robot could get enough experience to learn. If nothing the robot did helped it to win it would start trying random responses - just as we might try a 'wild and crazy idea' in the midst of a losing situation. Anybody on the list know anything about pattern recognition and how it could be implemented in euphoria for this kind of problem? I've seen some interesting stuff in other languages, but I'm afraid I'm limited to euphoria (or basic). I think some work in this arena would make our games top-notch. Last spring I worked on a program using a genetic algorithm to select animats which learned to search for food, avoid dangerous obstacles, eat, sleep, and breed in a limited environment. A working draft of the program was distributed to the list. But, although they learned, there was a ceiling to what they could learn and so I've abandoned that approach. ____________________________________________________________ T O P I C A -- Learn More. Surf Less. Newsletters, Tips and Discussions on Topics You Choose. http://www.topica.com/partner/tag01
2. Re: robot ai
- Posted by Kat <gertie at PELL.NET> Jan 30, 2001
- 466 views
On 30 Jan 2001, at 3:05, John McAdam wrote: > Anybody on the list know anything about pattern recognition and > how it could be implemented in euphoria for this kind of problem? > I've seen some interesting stuff in other languages, but I'm afraid > I'm limited to euphoria (or basic). Look at the code Graeme and David posted about findng the differences in two words, search the email archives for "global function diff". It's just a matter of changing MaxGap and applying the routine intellegently to your tokens, be they word representations or video bitmaps. Kat ____________________________________________________________ T O P I C A -- Learn More. Surf Less. Newsletters, Tips and Discussions on Topics You Choose. http://www.topica.com/partner/tag01
3. Re: robot ai
- Posted by Dan B Moyer <DanMoyer at PRODIGY.NET> Jan 30, 2001
- 484 views
Kat, Sorry I wasn't clearer about what I meant by "random" choices of action; what I meant was that random could be allowed for *worst* cases; in other words, the results of actions should be evaluated along a scale of relative success or failure, saved, and if the AI later found itself in a situation it did not have a "good" choice for, it could *then* try something at "random". Like, if you were in a *burning* building, and the door was blocked by something fallen outside it, and the room was 10 stories up with no fire escape, *then* it might in fact be useful if you "randomly" conceived to bash a hole in the floor to get down to the room below & try *it's* door. Dan ----- Original Message ----- From: "Kat" <gertie at PELL.NET> To: <EUforum at topica.com> Sent: Tuesday, January 30, 2001 10:00 AM Subject: RE: robot ai > On 30 Jan 2001, at 8:00, Chris Bensler wrote: > > > Can't say that I'm all that educated on this subject, but from what I > > know.. > > > > How would these bots 'remember' the patterns? Currently, with EU, the > > patterns would have to be saved to a loadable file.. > > Seems to me that is the best way to save data, retrieve a little data from massive > stores of it, and to spread the data to other creatures,,, the same way humans use > library books and email. > > > This is what I'm talking about.. the ability to have self modifying > > progs.. > > Programs that try RADICAL NEW concepts, devised by the AI.. not > > preprogrammed actions that can be combined to achieve a different result > > at RANDOM. > > I don't like this idea of "random" that's crept into the discussion. If i want to leave the > room, do i randomly try all methods? No, i never did that. I saw people passing thru > holes cut into the walls, called "doors" or "hallways" or in more than one case > "windows". Randomly deciding to exit by passing at a point averaged by the positions > of the "door" and the "window" didn't occur to me. And once i added the code to walk > on two legs, i still didn't try to get downstairs by tunneling directly thru the floor. I had > by then picked up a bit of code called "reasoning", and i reasoned if i made a hole in > the floor, this thing called "gravity" would eventually drop me thru it unexpectedly. > That's two examples of non-randomly adding and executing a segment of code. > > > Kat > > ____________________________________________________________ > T O P I C A -- Learn More. Surf Less. > Newsletters, Tips and Discussions on Topics You Choose. > http://www.topica.com/partner/tag01 > ____________________________________________________________ T O P I C A -- Learn More. Surf Less. Newsletters, Tips and Discussions on Topics You Choose. http://www.topica.com/partner/tag01
4. Re: robot ai
- Posted by Kat <gertie at PELL.NET> Jan 30, 2001
- 477 views
On 30 Jan 2001, at 15:53, Dan B Moyer wrote: > Kat, > > Sorry I wasn't clearer about what I meant by "random" choices of action; > what I meant was that random could be allowed for *worst* cases; in other > words, the results of actions should be evaluated along a scale of relative > success or failure, saved, and if the AI later found itself in a situation > it did not have a "good" choice for, it could *then* try something at > "random". Like, if you were in a *burning* building, and the door was > blocked by something fallen outside it, and the room was 10 stories up with > no fire escape, *then* it might in fact be useful if you "randomly" > conceived to bash a hole in the floor to get down to the room below & try > *it's* door. I still would not call that random, for you have reasoned that you could get out thru the door in the room below if you pursued the action of "bashing" thru the floor. You can't even say the program arrived at the sequence of bashing thru the floor and escaping by randomness in the program, unless you pre-programmed a BashThruFloor procedure that the program can execute. If humans could not execute vars, we'd all still be eating bugs and living in caves, because we don't have the genetic procedures LiveInHouse and EatBroiledCowWithPotatoesAndButterAndSaltOnChinaPlates. It's the reasoning part that has hung up Ai forever, "common sense" isn't easily duplicated, mostly, imho, the puter simply doesn't have enough knowledge, and/or doesn't have a smart enough programmer to use *unforseen* knowledge and occasions. That's why the ability to execute instructions that the programmer has not programmed, such as those instructions the puter has put into a variable, instruction sets that deal with a form of data unknown by the programmer when the program was written, is so important. Including the ability of the puter to override it's "instincts", it programming, as better data and forms of dealing with it, are thought of. Come on, Robert, reconsider the blanket "no", and give Eu a rare key to the future, a significant edge over other common general purpose languages. Lets execute variables! Kat > > Dan > > ----- Original Message ----- > From: "Kat" <gertie at PELL.NET> > To: <EUforum at topica.com> > Sent: Tuesday, January 30, 2001 10:00 AM > Subject: RE: robot ai > > > > On 30 Jan 2001, at 8:00, Chris Bensler wrote: > > > > > Can't say that I'm all that educated on this subject, but from what I > > > know.. > > > > > > How would these bots 'remember' the patterns? Currently, with EU, the > > > patterns would have to be saved to a loadable file.. > > > > Seems to me that is the best way to save data, retrieve a little data from > massive > > stores of it, and to spread the data to other creatures,,, the same way > humans use > > library books and email. > > > > > This is what I'm talking about.. the ability to have self modifying > > > progs.. > > > Programs that try RADICAL NEW concepts, devised by the AI.. not > > > preprogrammed actions that can be combined to achieve a different result > > > at RANDOM. > > > > I don't like this idea of "random" that's crept into the discussion. If i > want to leave the > > room, do i randomly try all methods? No, i never did that. I saw people > passing thru > > holes cut into the walls, called "doors" or "hallways" or in more than one > case > > "windows". Randomly deciding to exit by passing at a point averaged by the > positions > > of the "door" and the "window" didn't occur to me. And once i added the > code to walk > > on two legs, i still didn't try to get downstairs by tunneling directly > thru the floor. I had > > by then picked up a bit of code called "reasoning", and i reasoned if i > made a hole in > > the floor, this thing called "gravity" would eventually drop me thru it > unexpectedly. > > That's two examples of non-randomly adding and executing a segment of > code. > > > > > > Kat > > > > ____________________________________________________________ > > T O P I C A -- Learn More. Surf Less. > > Newsletters, Tips and Discussions on Topics You Choose. > > http://www.topica.com/partner/tag01 > > > > ____________________________________________________________ > T O P I C A -- Learn More. Surf Less. > Newsletters, Tips and Discussions on Topics You Choose. > http://www.topica.com/partner/tag01 > > ____________________________________________________________ T O P I C A -- Learn More. Surf Less. Newsletters, Tips and Discussions on Topics You Choose. http://www.topica.com/partner/tag01
5. Re: robot ai
- Posted by Dan B Moyer <DanMoyer at PRODIGY.NET> Jan 30, 2001
- 474 views
Kat, Don't need a BashThroughFloor routine, just a Bash routine, with destruction of object bashed, which program almost certainly already has. Don't need to *reason* it, just ability to pick some existing action and do it in problem circumstances and remember results; if it works, good; if not, bad. Assume randomly picked "bash" routine; bash what; what is there; see vase; bash vase, result: not exit fire; result not good; imperative: leave fire; see wall; bash wall; go new room; no exit; not good; see floor; bash floor; fall through; in new room; property: exit not known to be blocked; possible good; look for door; find exit; leave; good result; remember: bash floor if trapped with no exit. No reasoning involved, just search past evaluated results and allow random choice of action if no *good* choice found for circumstances. Dan ----- Original Message ----- From: "Kat" <gertie at PELL.NET> To: <EUforum at topica.com> Sent: Tuesday, January 30, 2001 4:31 PM Subject: Re: robot ai > On 30 Jan 2001, at 15:53, Dan B Moyer wrote: > > > Kat, > > > > Sorry I wasn't clearer about what I meant by "random" choices of action; > > what I meant was that random could be allowed for *worst* cases; in other > > words, the results of actions should be evaluated along a scale of relative > > success or failure, saved, and if the AI later found itself in a situation > > it did not have a "good" choice for, it could *then* try something at > > "random". Like, if you were in a *burning* building, and the door was > > blocked by something fallen outside it, and the room was 10 stories up with > > no fire escape, *then* it might in fact be useful if you "randomly" > > conceived to bash a hole in the floor to get down to the room below & try > > *it's* door. > > I still would not call that random, for you have reasoned that you could get out thru the > door in the room below if you pursued the action of "bashing" thru the floor. You can't > even say the program arrived at the sequence of bashing thru the floor and escaping by > randomness in the program, unless you pre-programmed a BashThruFloor procedure > that the program can execute. If humans could not execute vars, we'd all still be eating > bugs and living in caves, because we don't have the genetic procedures LiveInHouse > and EatBroiledCowWithPotatoesAndButterAndSaltOnChinaPlates. > > It's the reasoning part that has hung up Ai forever, "common sense" isn't easily > duplicated, mostly, imho, the puter simply doesn't have enough knowledge, and/or > doesn't have a smart enough programmer to use *unforseen* knowledge and > occasions. That's why the ability to execute instructions that the programmer has not > programmed, such as those instructions the puter has put into a variable, instruction > sets that deal with a form of data unknown by the programmer when the program was > written, is so important. Including the ability of the puter to override it's "instincts", it > programming, as better data and forms of dealing with it, are thought of. > > Come on, Robert, reconsider the blanket "no", and give Eu a rare key to the future, a > significant edge over other common general purpose languages. Lets execute variables! > > Kat > > > > > > Dan > > > > ----- Original Message ----- > > From: "Kat" <gertie at PELL.NET> > > To: <EUforum at topica.com> > > Sent: Tuesday, January 30, 2001 10:00 AM > > Subject: RE: robot ai > > > > > > > On 30 Jan 2001, at 8:00, Chris Bensler wrote: > > > > > > > Can't say that I'm all that educated on this subject, but from what I > > > > know.. > > > > > > > > How would these bots 'remember' the patterns? Currently, with EU, the > > > > patterns would have to be saved to a loadable file.. > > > > > > Seems to me that is the best way to save data, retrieve a little data from > > massive > > > stores of it, and to spread the data to other creatures,,, the same way > > humans use > > > library books and email. > > > > > > > This is what I'm talking about.. the ability to have self modifying > > > > progs.. > > > > Programs that try RADICAL NEW concepts, devised by the AI.. not > > > > preprogrammed actions that can be combined to achieve a different result > > > > at RANDOM. > > > > > > I don't like this idea of "random" that's crept into the discussion. If i > > want to leave the > > > room, do i randomly try all methods? No, i never did that. I saw people > > passing thru > > > holes cut into the walls, called "doors" or "hallways" or in more than one > > case > > > "windows". Randomly deciding to exit by passing at a point averaged by the > > positions > > > of the "door" and the "window" didn't occur to me. And once i added the > > code to walk > > > on two legs, i still didn't try to get downstairs by tunneling directly > > thru the floor. I had > > > by then picked up a bit of code called "reasoning", and i reasoned if i > > made a hole in > > > the floor, this thing called "gravity" would eventually drop me thru it > > unexpectedly. > > > That's two examples of non-randomly adding and executing a segment of > > code. > > > > > > > > > Kat > > > > ____________________________________________________________ T O P I C A -- Learn More. Surf Less. Newsletters, Tips and Discussions on Topics You Choose. http://www.topica.com/partner/tag01
6. Re: robot ai
- Posted by Kat <gertie at PELL.NET> Jan 30, 2001
- 473 views
On 30 Jan 2001, at 16:57, Dan B Moyer wrote: > Kat, > > Don't need a BashThroughFloor routine, just a Bash routine, with destruction > of object bashed, which program almost certainly already has. Will you code *every* verb, and then *all* the attributes of every noun the verbs use and affect, with adjectives and adverbs, including all prepositions and pronouns? Have you looked into Cyc lately? Ouch. What if the bot is sent to space by NASA? I can't do that alone, and it's not learning to have the programmers try to foresee every activity and permutation. Kat >Don't need to > *reason* it, just ability to pick some existing action and do it in problem > circumstances and remember results; if it works, good; if not, bad. Assume > randomly picked "bash" routine; bash what; what is there; see vase; bash > vase, result: not exit fire; result not good; imperative: leave fire; see > wall; bash wall; go new room; no exit; not good; see floor; bash floor; fall > through; in new room; property: exit not known to be blocked; possible good; > look for door; find exit; leave; good result; remember: bash floor if > trapped with no exit. No reasoning involved, just search past evaluated > results and allow random choice of action if no *good* choice found for > circumstances. > > Dan > > > > ----- Original Message ----- > From: "Kat" <gertie at PELL.NET> > To: <EUforum at topica.com> > Sent: Tuesday, January 30, 2001 4:31 PM > Subject: Re: robot ai > > > > On 30 Jan 2001, at 15:53, Dan B Moyer wrote: > > > > > Kat, > > > > > > Sorry I wasn't clearer about what I meant by "random" choices of action; > > > what I meant was that random could be allowed for *worst* cases; in > other > > > words, the results of actions should be evaluated along a scale of > relative > > > success or failure, saved, and if the AI later found itself in a > situation > > > it did not have a "good" choice for, it could *then* try something at > > > "random". Like, if you were in a *burning* building, and the door was > > > blocked by something fallen outside it, and the room was 10 stories up > with > > > no fire escape, *then* it might in fact be useful if you "randomly" > > > conceived to bash a hole in the floor to get down to the room below & > try > > > *it's* door. > > > > I still would not call that random, for you have reasoned that you could > get out thru the > > door in the room below if you pursued the action of "bashing" thru the > floor. You can't > > even say the program arrived at the sequence of bashing thru the floor and > escaping by > > randomness in the program, unless you pre-programmed a BashThruFloor > procedure > > that the program can execute. If humans could not execute vars, we'd all > still be eating > > bugs and living in caves, because we don't have the genetic procedures > LiveInHouse > > and EatBroiledCowWithPotatoesAndButterAndSaltOnChinaPlates. > > > > It's the reasoning part that has hung up Ai forever, "common sense" isn't > easily > > duplicated, mostly, imho, the puter simply doesn't have enough knowledge, > and/or > > doesn't have a smart enough programmer to use *unforseen* knowledge and > > occasions. That's why the ability to execute instructions that the > programmer has not > > programmed, such as those instructions the puter has put into a variable, > instruction > > sets that deal with a form of data unknown by the programmer when the > program was > > written, is so important. Including the ability of the puter to override > it's "instincts", it > > programming, as better data and forms of dealing with it, are thought of. > > > > Come on, Robert, reconsider the blanket "no", and give Eu a rare key to > the future, a > > significant edge over other common general purpose languages. Lets execute > variables! > > > > Kat > > > > > > > > > > Dan > > > > > > ----- Original Message ----- > > > From: "Kat" <gertie at PELL.NET> > > > To: <EUforum at topica.com> > > > Sent: Tuesday, January 30, 2001 10:00 AM > > > Subject: RE: robot ai > > > > > > > > > > On 30 Jan 2001, at 8:00, Chris Bensler wrote: > > > > > > > > > Can't say that I'm all that educated on this subject, but from what > I > > > > > know.. > > > > > > > > > > How would these bots 'remember' the patterns? Currently, with EU, > the > > > > > patterns would have to be saved to a loadable file.. > > > > > > > > Seems to me that is the best way to save data, retrieve a little data > from > > > massive > > > > stores of it, and to spread the data to other creatures,,, the same > way > > > humans use > > > > library books and email. > > > > > > > > > This is what I'm talking about.. the ability to have self modifying > > > > > progs.. > > > > > Programs that try RADICAL NEW concepts, devised by the AI.. not > > > > > preprogrammed actions that can be combined to achieve a different > result > > > > > at RANDOM. > > > > > > > > I don't like this idea of "random" that's crept into the discussion. > If i > > > want to leave the > > > > room, do i randomly try all methods? No, i never did that. I saw > people > > > passing thru > > > > holes cut into the walls, called "doors" or "hallways" or in more than > one > > > case > > > > "windows". Randomly deciding to exit by passing at a point averaged by > the > > > positions > > > > of the "door" and the "window" didn't occur to me. And once i added > the > > > code to walk > > > > on two legs, i still didn't try to get downstairs by tunneling > directly > > > thru the floor. I had > > > > by then picked up a bit of code called "reasoning", and i reasoned if > i > > > made a hole in > > > > the floor, this thing called "gravity" would eventually drop me thru > it > > > unexpectedly. > > > > That's two examples of non-randomly adding and executing a segment of > > > code. > > > > > > > > > > > > Kat > > > > > > > > > ____________________________________________________________ > T O P I C A -- Learn More. Surf Less. > Newsletters, Tips and Discussions on Topics You Choose. > http://www.topica.com/partner/tag01 > > ____________________________________________________________ T O P I C A -- Learn More. Surf Less. Newsletters, Tips and Discussions on Topics You Choose. http://www.topica.com/partner/tag01