DEVS-based Spatiotemporal Gating Model for Controlling LLM Agent Invocations in Spatial Computing Environments

In spatial computing (XR) environments, simple distance-based LLM agent triggers often cause pass-by false positives, boundary chattering, and non-deterministic behavior. This paper proposes a PDEVS-based SpatialGatingAgent to address these issues through formal spatio-temporal gating. The model formalizes gating rules into explicit time and concurrency semantics, using the time advance function (ta) for dwell-time management and the confluent transition function (δ_conf) to eliminate execution-order dependency during simultaneous events. Simulation results across five scenarios (S1– S5) demonstrate that the proposed method ensures reliable invocation control consistent with the gating policy by suppressing unnecessary calls and guaranteeing deterministic outcomes specifically in con- fluent edge cases. These results suggest that the proposed formalization provides a robust framework for consistent agent interaction in complex XR environments.