Steven.Minton@CAD.CS.CMU.EDU (04/19/88)
Ken, Can you post the following on AIList? Thanks. I recently saw a note on AIList (v6 #52) from Vibhu Mittal asking whether there are any planning systems written in vanilla CommonLisp that are available for educational purposes. I think that the PRODIGY system built here at CMU would probably be well-suited to his purposes. PRODIGY is a domain-independent problem solving system based on the STRIPS architecture (with some interesting improvements). We use it as a testbed for machine learning research (for example, see articles in IJCAI-87 and the '85 and '87 machine learning workshops). There are several hot issues in planning that the current PRODIGY implementation does not deal with (such as reasoning about uncertainty, temporal constraints, conditional plans). So if you are looking for a planner that is state of the art in all respects, PRODIGY is probably not what you are looking for. On the other hand, it is a relatively elegant and powerful system that extands the well-understood STRIPS approach to planning. So it's just fine if you're primarily interested in ML issues (especially if you want to be able to compare your work to previous work in AI). I think it would also be perfect for educational purposes. Although PRODIGY allows you to specify surprisingly complex domains, if you want, you can simply load in the old blocksworld or STRIPS domains. Students can get there hands on a running system, and see the advantages and limitations of this approach to planning. The vanilla PRODIGY 1.0 problem solver was recently released for external use, and comes with a manual and a few example task domains. An interactive trace facility and graphics capibilities are included. The system's description language is based on predicate calculus (and includes conjunction, disjunction, and existential and universal quantification). The user can write explicit search control rules to guide the problem solver's search. The system is free and can be FTPed from CMU. PRODIGY 2.0 is expected to be released this summer (no guarantees, I still have yet to complete the port from Franzlisp), and will include an advanced EBL learning system (as was described in my thesis). Later versions are expected to include an automatic abstraction mechanism, a learning-by-experimentation module, a derivational analogy module, an interface to the CMU World modelers system, and a full extension of the learning apprentice interface. These are all current research projects at CMU. Send requests to snm@cs.cmu.edu. Allow some time for me to get back to you, I'm trying to get some real work done too. (I can't promise anything in the way of support.) - Steve Minton (& Craig Knoblock)