GUIDE-EX: A Declarative Execution Architecture for Hierarchical Robotic Task Execution
Abstract
The integration of modern artificial intelligence into robotic manipulation is consistently hindered by the structural gap between rigid industrial automation standards and the abstract requirements of learning-based policies. This paper introduces GUIDE Execution (GUIDE-EX), a declarative, hardware-agnostic execution architecture derived from the Laboratory Automation Plug & Play (LAPP) architecture. The framework introduces three structural primitives—Actions, Observations, and an orthogonal Utility dimension—to provide a polymorphic foundation that uniformly supports classical deterministic sequences, cognitive hybrid architectures, and Vision-Language-Action (VLA) models. Translating this theoretical flexibility into practice, the architecture was validated in a simulated, cluttered pick-and-place scenario. The results demonstrate that the declarative framework itself significantly reduces operational complexity, lowering top-level orchestration overhead from 35 programmatic steps to 11 configuration nodes. Furthermore, the implemented error recovery mechanism improves synthetic data generation efficiency—requiring 114 attempts for 100 successful episodes compared to a 149-attempt baseline—while inherently enriching datasets with vital error-recovery trajectories for robust Sim-to-Real transfer.