This Julia package implements the incremental pruning solver for partially observable Markov decision processes.

using Pkg
Pkg.add("IncrementalPruning")

using IncrementalPruning
using POMDPModels
pomdp = TigerPOMDP() # initialize POMDP

solver = PruneSolver() # set the solver

policy = solve(solver, pomdp) # solve the POMDP  

The result of solve is a Policy that contains the alpha vectors of the solution.

IncrementalPruning.jl solves problems implemented using the POMDPs.jl interface. See the documentation for POMDPs.jl for more information.

This solver implements the incremental pruning algorithm as described in Zhang and Liu (1996) and Cassandra et al. (1997). This solution method is exact (ϵ-optimal) but is much slower than modern approximate solution techniques. As such, it is only computationally feasible for small problems.

Cassandra, A., Littman, M., & Zhang, N. (1997). Incremental pruning: A simple, fast, exact method for partially observable Markov decision processes. Proceedings of the Thirteenth Annual Conference on Uncertainty in Artificial Intelligence (UAI-97), 54–61.

Zhang N. L., Liu W. (1996). Planning in stochastic domains: Problem characteristics and approximation. Technical Report HKUST-CS96-31, Hong Kong University of Science and Technology.