1. Christian Muise
  2. Planner for Relevant Policies

Wiki

Clone wiki

Planner for Relevant Policies / Prob-PRP

Prob-PRP: Using PRP for Solving Probabilistic Planning Problems

Prob-PRP is a version of PRP that can parse and solve many PPDDL problems [1]. You can get access to Prob-PRP by checking out this branch. To run, use src/prp as normal, except with PPDDL files in place of FOND ones.

Benchmarks used in the publication can be found in the prob-benchmarks directory.

Current Caveats

  • While Prob-PRP aims to maximize the probability of achieving the goal, it is only guaranteed to do so if a policy exists with 100% success rate.
  • While Prob-PRP can parse many PPDDL problem files, it cannot optimize a reward function. It is thus definitely not a MAXREWARD planner, but instead a MAXPROB planner.

[1] The wrong mix of probabilistic and conditional effects can cause a problem. Generally you are safe if the probabilities are all at the outer-most level of the action effects.

Updated