I am learning classical planning techniques, and I know many examples of robots are in books. However, I am more curious about the real-world usage of these techniques.
Are PDDL and classical planners (e.g. FF) directly used in industry or consumer products? Or does every company or organization have its own representation of problems rather than directly using PDDL as representation language?
Rarely. PDDL is just one way of specifying search problems. First, there are other modeling languages, for example from the model-checking community, such as SMV, NuSMV, and a number of others, which are equally well (and sometimes better) suited for specifying planning and other equivalent state-space search and reachability problems.
Second, when a company needs to solve a specific planning problem, the most scalable way of doing that is to not use domain-independent planners and modeling languages such as PDDL, but to develop problem-specific heuristics and pruning methods, and apply them without having to go through a modeling language. In many cases this is not much more work than writing a specification in PDDL, especially considering that PDDL is a relatively low level specification language, with little support to modeling complex domains.
PDDL has, for example, only very limited ways of using numbers and arithmetic. Real-world planning and scheduling is very much a disjoint thing from the AI planning activities in the academia. There is hardly any connection between them. PDDL is very rarely used in the industry.
