Recherche incrémentale à base de points pour la résolution des DEC-POMDPs (présentation courte)

Recent scaling up of decentralized partially observable Markov decision process (DECPOMDP) solvers towards realistic applications is mainly due to approximate methods. Of this family, MEMORY BOUNDED DYNAMIC PROGRAMMING (MBDP), which combines in a suitable manner top-down heuristics and bottomup value function updates, can solve DECPOMDPs with large horizons. The performance of MBDP, however, can be drastically improved by avoiding the systematic generation and evaluation of all possible policies which result from the exhaustive backup. To achieve that, we suggest a heuristic search method, namely POINT BASED INCREMENTAL PRUNING (PBIP) which is able to distinguish policies with different heuristic estimate. Taking this insight into account, PBIP searches only among the most promising policies, finds the useful, and prunes dominated ones. Doing so permits us to reduce clearly the amount of computation required by the exhaustive backup. A theoretical analysis of PBIP shows that it is both complete and optimal (with respect to MBDP). We also present experiment results, which show how PBIP outperforms the performance of MBDP and extensions on DECPOMDP examples from the literature.