|
Is there any software that acts as an intersection between contemporary OWL/RDF reasoners, and the older STRIPS-style automated planners and schedulers? Both systems make use of RETE-based pattern matching, but only the automated planners seem to formalise the concept of an "action". Unfortunately, all the projects I've found that implemented automated planning, like Graphplan or SOAR, seem to be dead or dying, and never seemed to scale well to begin with. Some appear to have been inactive for nearly 10 years. Current data stores are implemented on RDMS and can scale to and reason over millions of triples, but I haven't found any that specifically try and reason over actions. I can envision how the concept of actions might be represented in traditional RDF, but I'm sure it would still be very complicated and hackish without official support. Unfortunately, I can't find much prior art. Has this been done before? |
|
Planners have been improving but you are right that most still do not scale too well. There has been some work previously on "Description Logic + AI Planning" (try google search on this), but I am not aware of much active research on this topic, or usages of such planners. In general, I see very little real-life applications of AI Planning outside of Mars Rovers and other space applications, and now a bit in computer games. It would be interesting to understand some specific use-case ideas you have in mind. Personally I had exactly the same question as you ask in my mind before (I mean about OWL + Planning) and am also interested in this problem in the long term. I've had difficulty thinking of practical use cases as well, since so many "actions" for practical problems aren't captured digitally. However, one idea I had was for an automated systems administrator, that could formulate plans for configuring servers and/or repair or workaround common problems.
(Feb 24 '11 at 13:46)
Cerin
By "Description Logic + AI Planning", were you referring to HTN-DL?
(Feb 24 '11 at 20:28)
Cerin
@Cerin Yes, HTN-DL is one approach. Also see "Description Logics" at http://www.isi.edu/~blythe/cs541/
(Feb 25 '11 at 23:11)
Amit Kumar
|
|
Don't know if this is what you're looking for: HotPlanner. I had stumbled across this myself. It does look very close to what I'm looking for. Unfortunately, it's closed source proprietary licensing disqualifies it.
(Mar 06 '11 at 23:00)
Cerin
|
If you mean that Soar is not being developed in the direction of your interest, then while you are right, it was a very unfortunate way to put it.
@Lukasz, How do you mean? I only meant that Soar's been around for over 20 years, but still has a huge learning curve, a very small userbase and no public examples of practical applications. Please correct me if I'm wrong about this.
@Cerin, we are close to matters of opinion rather than of fact. I don't agree that Soar has a steep learning curve -- on the contrary, the software is very usable (for toy problems). The user base is very small (as for a system of such fame), but I think it is stable, they are holding annual workshops. I agree that Soar's achievements look pale given how long the system has been in development. I would agree/argue that the initial SOAR cognitive architecture is not the best one out there, and now the Soar community is experimenting around that. But going down the description logic path would take them even further away from a sound cognitive architecture.
My initial point was that in absolute terms, a project that holds annual workshops of sustained quality is not dying.
@Lukasz, Those are fair points. What cognitive architectures do you prefer over Soar?