The system is centrally controlled. As cars begin building up in one are of the system, empty cars are given commands to reroute them to empty areas of the system. Because the stations are on separate tracks from, you can have queues of empty cars in the station. Load the front one, send it on its way, and then load the next one. Multiple cars could be loaded at once during heavier use times. The cars themselves are really cheap and don't use power when they are not running. There is not much of a drawback to ahving many more cars than the system use requires.
Not to mention the fact that busses come in 10-40 minute intervals (or so). You do not have to wait with this system. Presumably, there will be empty cars at each stop waiting for passengers.
If you have id cards to let you in then you know who used that car. If enough cars used by that individual are vandalized, then that person could be banned from the system (or penalized in some way). If cameras are installed in cars, then when the suspected individual enters the vehicle a monitor turns on at the home office to ensure that the person is acutally doing damage. I suspect that the knowledge that vandalizers will be caught will decrease the incidents of vandalism.
However, use of id cards also means the governement can track your movements and movement habits.
Aibos are much more like real robots than the Roombas. The charger location on the roomba isn't very sophisticated. It senses an infrared signal emitted by the charger and follows reactive control laws that maximize the signal -- much like a moth to flame.
The gradparent is right. Roomba does not do any sort of mapping to figure out where it hasn't cleaned or where it is okay to travel. It is much more similar to a bump-and-go toy car than what most people think of as a robot. This doesn't change the fact that it is a nice designed and commercially successful product.
My guess is that you will be able to subscribe to HBO and other pay channels directly through XBox live within a few years. No more cable packages full of channels you have no interest in.
I like the unix functionality on mac osx. cygwin doesn't cut it for me. I like the OSX UI more than the free window managers. Beside, I doubt Windows on the mac would have very much support from software vendors.
In a group playing non-iterated PD, a defector playing against a cooperator will beat a cooperator playing against a defector. I still think defecting is the correct strategic plan (which isn't how I run my life since these things are usually iterative -- often unexpectedly so).
An interesting note, when designing the PD payoffs, if the payoffs are such that CD+DC > 2*CC, then it is in both agents' interests to swap between coop and defect out of synch with the other.
One factor that might have hurt TFT in this tournament is the addition of noise. If TFT thinks the opponent cheated, even if they didn't, TFT will cheat on the next turn. If it is a TFT vs TFT situation, both players will alternately cooperate/defect.
I think this winning team's solution was interesting. Having the ability to recognize a cheater so that you can punish them is necessary for the coop/coop solution. I've heard it speculated that one reason humans have evolved distinctive faces is to make it easier to distinguish social cooperators from cheaters. I personally believe that a lot of the social problems in society today stem from the fact that we live in a mostly anonymous society.
I wrote about some of this last year in my one and only./ journal entry
Unfortunately, PD has a dominant strategy to cheat your opponent. No matter what your oppoent does, you can do better by cheating.
Iterated PD is more interested since it lets your opponent punish you for cheating. So you get into some interesting social issues.
How well a given strategy does depends on the strategies other in its community are using. If the population is heavily cheater based, then agents that cooperate will lose big time. However, if there are enough cooperators, then they will form a coalition of sorts, and even though they will lose to the cheaters, in the end they will come out on top.
The software is supposed to handle "reconfiguration". If a component fails, the software can direct the system to work around the failed component so that the vehicle is still operational. I'm pretty sure that the Livingstone model has to be written to know how to work around a failed component. But with a proper model, it should be quite capable of determining when a component is broken, work around it, and continue the mission without human intervention.
Not so important for a satellite, but very important for a deep space mission when communication lag to Earth could result in unrecoverable failures.
We were using Adnan Darwiche's papers. I don't have any figures on component count vs diagnosis time. I did some porting from Lisp to C for DNNF structure evaluation (for some reason C is more trusted than Lisp on autonomous spacecraft), and worked on building models for interferometry telescopes.
I could probably ping my manager and get some data on the model sizes if you are interested.
I worked on a project very similar to this at JPL a little over a year ago. You specify the entire system in a language. We then coverted the specification into a large conjunctive normal form (CNF) boolean statement. The CNF form was converted into something called decomposable negagation normal form (DNNF) which lets you find satisfiable interpretations in linear time with respect to the DNNF structure. You assign sensor readings and expected states to the terminals of the DNNF tree and then do SAT to find other possible values. You can also give different possible sensor values and states different weights to find the most probable (least cost) state. The compilation process takes a long time, but you only have to do it once and you have a known space and time process to upload to the vehicle.
It is interesting that once an AI problem is solved, most people stop regarding it as AI. A* search was once at the cutting edge of AI.
I worked on a project very similar to this at JPL a little over a year ago. You specify the entire system in a language. We then coverted the specification into a large conjunctive normal form (CNF) boolean statement. The CNF form was converted into something called decomposable negagation normal form (DNNF) which lets you find satisfiable interpretations in linear time with respect to the DNNF structure. You assign sensor readings and expected states to the terminals of the DNNF tree and then do SAT to find other possible values. You can also give different possible sensor values and states different weights to find the most probable (least cost) state. The compilation process takes a long time, but you only have to do it once and you have a known space and time process to upload to the vehicle.
In the more recent "Ultimate Spider-man", the black suit was a joint project of Peter's and Eddie Brock's fathers to find a cure for cancer. Brock is a graduate student who is still researching the cure. Peter gets a little bit on his finger and it covers his whole body. He has the suit for a few issues (which lets him change clothes, shoot webbing, and heal from bullet wounds) but realizes that it is making him more violent so he gets rid of it and tries to destroy the rest of Eddie's supply. Eddie has a secret stash and tried to replicate Peter's success. The suit nearly eats him, but he emerges insane as Venom with a hatred of Spider-man for trying to destroy his father's work.
Slashdot Mirror
The system is centrally controlled. As cars begin building up in one are of the system, empty cars are given commands to reroute them to empty areas of the system. Because the stations are on separate tracks from, you can have queues of empty cars in the station. Load the front one, send it on its way, and then load the next one. Multiple cars could be loaded at once during heavier use times. The cars themselves are really cheap and don't use power when they are not running. There is not much of a drawback to ahving many more cars than the system use requires.
Not to mention the fact that busses come in 10-40 minute intervals (or so). You do not have to wait with this system. Presumably, there will be empty cars at each stop waiting for passengers.
If you have id cards to let you in then you know who used that car. If enough cars used by that individual are vandalized, then that person could be banned from the system (or penalized in some way). If cameras are installed in cars, then when the suspected individual enters the vehicle a monitor turns on at the home office to ensure that the person is acutally doing damage. I suspect that the knowledge that vandalizers will be caught will decrease the incidents of vandalism.
However, use of id cards also means the governement can track your movements and movement habits.
It IS in the shape of a sneaker. It's even the same size. Do you think that everyone in the galaxy has feet shaped like we roadblocks do?
Actually, this is the governement's new approach to drug control. Let the private sector handle it.
Those 'liberals' aren't too far off mark about the guns bit:
http://www.shotspotter.com/index.shtml
Aibos are much more like real robots than the Roombas. The charger location on the roomba isn't very sophisticated. It senses an infrared signal emitted by the charger and follows reactive control laws that maximize the signal -- much like a moth to flame.
The gradparent is right. Roomba does not do any sort of mapping to figure out where it hasn't cleaned or where it is okay to travel. It is much more similar to a bump-and-go toy car than what most people think of as a robot. This doesn't change the fact that it is a nice designed and commercially successful product.
My guess is that you will be able to subscribe to HBO and other pay channels directly through XBox live within a few years. No more cable packages full of channels you have no interest in.
I like the unix functionality on mac osx. cygwin doesn't cut it for me. I like the OSX UI more than the free window managers. Beside, I doubt Windows on the mac would have very much support from software vendors.
In a group playing non-iterated PD, a defector playing against a cooperator will beat a cooperator playing against a defector. I still think defecting is the correct strategic plan (which isn't how I run my life since these things are usually iterative -- often unexpectedly so).
An interesting note, when designing the PD payoffs, if the payoffs are such that CD+DC > 2*CC, then it is in both agents' interests to swap between coop and defect out of synch with the other.
One factor that might have hurt TFT in this tournament is the addition of noise. If TFT thinks the opponent cheated, even if they didn't, TFT will cheat on the next turn. If it is a TFT vs TFT situation, both players will alternately cooperate/defect.
./ journal entry
I think this winning team's solution was interesting. Having the ability to recognize a cheater so that you can punish them is necessary for the coop/coop solution. I've heard it speculated that one reason humans have evolved distinctive faces is to make it easier to distinguish social cooperators from cheaters. I personally believe that a lot of the social problems in society today stem from the fact that we live in a mostly anonymous society.
I wrote about some of this last year in my one and only
Unfortunately, PD has a dominant strategy to cheat your opponent. No matter what your oppoent does, you can do better by cheating.
Iterated PD is more interested since it lets your opponent punish you for cheating. So you get into some interesting social issues.
How well a given strategy does depends on the strategies other in its community are using. If the population is heavily cheater based, then agents that cooperate will lose big time. However, if there are enough cooperators, then they will form a coalition of sorts, and even though they will lose to the cheaters, in the end they will come out on top.
My one and only journal entry is about Prisoner's Dilemma (that I wrote during a class on auctions and game theory).
http://slashdot.org/~naoursla/journal
Alas, it is not. :(
Where are you working?
The software is supposed to handle "reconfiguration". If a component fails, the software can direct the system to work around the failed component so that the vehicle is still operational. I'm pretty sure that the Livingstone model has to be written to know how to work around a failed component. But with a proper model, it should be quite capable of determining when a component is broken, work around it, and continue the mission without human intervention.
Not so important for a satellite, but very important for a deep space mission when communication lag to Earth could result in unrecoverable failures.
We were using Adnan Darwiche's papers. I don't have any figures on component count vs diagnosis time. I did some porting from Lisp to C for DNNF structure evaluation (for some reason C is more trusted than Lisp on autonomous spacecraft), and worked on building models for interferometry telescopes.
I could probably ping my manager and get some data on the model sizes if you are interested.
I worked on a project very similar to this at JPL a little over a year ago. You specify the entire system in a language. We then coverted the specification into a large conjunctive normal form (CNF) boolean statement. The CNF form was converted into something called decomposable negagation normal form (DNNF) which lets you find satisfiable interpretations in linear time with respect to the DNNF structure. You assign sensor readings and expected states to the terminals of the DNNF tree and then do SAT to find other possible values. You can also give different possible sensor values and states different weights to find the most probable (least cost) state. The compilation process takes a long time, but you only have to do it once and you have a known space and time process to upload to the vehicle.
It is interesting that once an AI problem is solved, most people stop regarding it as AI. A* search was once at the cutting edge of AI.
I worked on a project very similar to this at JPL a little over a year ago. You specify the entire system in a language. We then coverted the specification into a large conjunctive normal form (CNF) boolean statement. The CNF form was converted into something called decomposable negagation normal form (DNNF) which lets you find satisfiable interpretations in linear time with respect to the DNNF structure. You assign sensor readings and expected states to the terminals of the DNNF tree and then do SAT to find other possible values. You can also give different possible sensor values and states different weights to find the most probable (least cost) state. The compilation process takes a long time, but you only have to do it once and you have a known space and time process to upload to the vehicle.
I worked on a project similar to this for NASA's interferometry telescopes at JPL a summer ago.
It is a bit funny, though, that this evolution takes the form of borrowing stuff from an ancient language.
Yeah! Its strange how all languages seem to converge towards Lisp.
In the more recent "Ultimate Spider-man", the black suit was a joint project of Peter's and Eddie Brock's fathers to find a cure for cancer. Brock is a graduate student who is still researching the cure. Peter gets a little bit on his finger and it covers his whole body. He has the suit for a few issues (which lets him change clothes, shoot webbing, and heal from bullet wounds) but realizes that it is making him more violent so he gets rid of it and tries to destroy the rest of Eddie's supply. Eddie has a secret stash and tried to replicate Peter's success. The suit nearly eats him, but he emerges insane as Venom with a hatred of Spider-man for trying to destroy his father's work.
You should put this post in your journal. I think it addresses the core belief differences between you and those who disagree with you.
Doesn't just about every game on the gameboy represent prior art in handhelds?
The longer I press the button, the higher Mario jumps.