Constraint-Bounded Logical Inference in Hybrid Neuro-Symbolic AI

Abstract

Hybrid neuro-symbolic reasoning introduces the ability to combine adaptive neural policy learning
with explicit logical constraints required in industrial robotic environments. This article presents a
constraint-bounded inference framework that positions a symbolic rule layer between neural action
proposals and the robotic actuation pipeline, ensuring that decision-making remains interpretable,
safe, and operationally feasible during continuous production workflows. The results show that the
system maintains task continuity under constraint pressure, prevents unsafe behavior in dynamic or
partially observable conditions, and enables rapid updates to operational rules without retraining
neural models. The architecture also improves transparency by producing human-readable rationale
for action arbitration, supporting industrial audit and supervisory requirements. The findings
demonstrate that constraint-bounded logical inference provides a scalable foundation for safe and
reliable autonomous robotics in dynamic manufacturing scenarios.