Comparative scalability of UMOP’s NADL/OBDD encoding versus MBP’s encoding

Determine how the OBDD-based encoding used in UMOP—specifically the NADL-based, partitioned transition-relation representation—scales and performs relative to the encoding used in MBP, including assessing whether MBP’s monolithic transition-relation encoding may outperform or match UMOP in domains where partitioning is unavailable and when both systems employ monolithic representations.

Background

UMOP introduces NADL and a partitioned OBDD-based transition-relation encoding intended to improve scalability, while MBP uses an encoding that, at the time of writing, does not support partitioning. The authors note comparable performance in the beam walk domain where both systems effectively use monolithic representations, raising questions about the practical advantages of partitioning across different domains.

Understanding the comparative scalability and performance characteristics of these encodings is crucial for selecting appropriate universal planning tools, especially in large, non-deterministic, multi-agent domains where OBDD operations and transition-relation structure dominate computational cost.

References

Our research has drawn our attention to a number of open questions that we would like to address in the future. In particular we wonder how well our encoding of planning problems scales compared to the encoding used by MBP. Currently MBP's encoding does not support a partitioned representation of the transition relation, but the encoding may have other properties that, despite the monolithic representation, may make it a better choice. The two systems may also have an equal performance when both are using a monolithic representation (as in the beam walk example), which should give UMOP an advantage in domains where a partitioning of the transition relation can be defined.

— OBDD-based Universal Planning for Synchronized Agents in Non-Deterministic Domains (1106.0229 - Jensen et al., 2011) in Section 8, Conclusion and Future Work