Several RockSample problems are included my ZMDP package. They are found in the src/pomdpModels directory, in the form of Perl scripts. Run the Perl script to generate the model in Tony Cassandra's POMDP model format. Just as a warning--the biggest model, RockSample_10_10.pomdp, has a file size of 180 MB and may take a few minutes to generate.
Download problems (ZMDP package) Described in our UAI 2004 paper
Provided as Cassandra-format POMDP models and "source files" that include human-readable maps. The POMDP models are flat models produced after CIVA abstraction. You can also find problem-specific code for reading a LifeSurvey source file, performing CIVA abstraction, and generating a POMDP model in the ZMDP package; look in the src/pomdpModels/lifeSurvey directory. (Unfortunately, it is not very well documented.)
Tony's pomdp-solve package is probably the best known POMDP software. Here at CMU it was used for some years as a teaching tool. It implements several exact value iteration algorithms that today you could call "classics". My ZMDP package uses some code from pomdp-solve for parsing POMDP models (Tony developed the model format that is most widely used today).
My sense from a few years ago is that pomdp-solve gets good marks for ease of use and stability, but the algorithms it implemented were not competitive with more recent algorithms in terms of quick approximate solutions to large POMDPs. But that may have changed. I notice there is now a version of PBVI called "Finite Grid" in there, which I haven't tried out.
Perseus is a well-known point-based value iteration algorithm. Since the original paper was published, the algorithm has been extended to use an ADD model representation ("Symbolic Perseus", see below) and also to work for certain continuous POMDPs. I should also mention that Matthijs is a nice guy (I had a few chances to meet him when he spent a semester visiting at CMU). So far I haven't actually used this package... I am interested to hear experiences from others.
This package sounds like a good one. Again, I haven't tried it out yet. It combines basic Perseus with ADD data structures to compactly represent large factored POMDPs. The solver been used for some interesting applications in elder care.
Tarek wrote this parser for his own work and would generously like to make it available for other people using ZMDP. I haven't tried it out myself.