Smart Robotic Manufacturing Operations with Adaptive Task Planning: GPT-Augmented Multi-Channel Ensemble Learning with Situational Awareness

This paper presents a smart robotic operations framework that integrates multi-camera perception, GPT-augmented multi-channel ensemble learning, and domain-driven task planning for dynamic manufacturing environments. By leveraging synchronized top, side, and scene cameras, the system achieves high-precision object localization and real-time situation awareness across complex work zones. Stacking-style ensemble learning fuse multi-view pose estimates into reliable robot control commands that adapt to variable spatial conditions. GPT- augmented reasoning further enables automatic generation of Planning Domain Definition Language (PDDL) models, which, together with ROSPlan integration, translate high-level task plans into executable robot actions. A preliminary use case in flexible electronics component assembly demonstrates the system’s capability to handle multi-task scheduling, material changeover coordination, and task prioritization across parallel workstations. This framework bridges perception, reasoning, and manipulation within a closed-loop architecture, offering a scalable approach toward adaptive, situation-aware robotic operations in manufacturing. Future work will focus on real-world deployment, enhanced domain modeling, and broader integration with production management systems.