JuliaPOMDP/PointBasedValueIteration.jl

Continuous Observations

Opened this issue · 1 comments

It would be nice to have an offline solver, which can handle continuous observation spaces (or large discrete ones). Hoey & Poupart [1] introduced an extension of point-based methods, which works by partitioning the observation space into a small set of regions, which lead to the same conditional plan.

I might give implementing their method a try at some point, but I am not sure whether this should be implemented as an extension of PointBasedValueIteration.jl or as a standalone package.

[1] Hoey, J., & Poupart, P. (2005, July). Solving POMDPs with continuous or large discrete observation spaces. In IJCAI (pp. 1332-1338).

This would be really cool. I'm guessing we would want a more convenient way to represent continuous spaces though. If it is a pretty straightforward addition, I think it would make sense to be in this package, but it could also just depend on this package without being a part of it.