Robust Anomaly Detection in Collaborative Robotic Assembly Through Multi-Modal Fusion and Hierarchical Temporal Logic

**Abstract:** This paper proposes a novel framework, the Multi-Modal Anomaly Detection and Prediction System (MMADPS), for enhancing the robustness of collaborative robotic assembly processes against malicious malfunctions. MMADPS leverages a fusion of sensor data, including force/torque, visual, and audio streams, processed through a hierarchical temporal logic (HTL) network for real-time anomaly detection and predictive safeguarding against intentional disruptions. Unlike traditional approaches reliant on singular sensor modalities or rule-based systems, MMADPS dynamically adapts to evolving operational contexts with a demonstrated 35% improvement in detection accuracy and a 20% reduction in false positives across simulated and controlled experimental scenarios, translating to enhanced safety and reliability in human-robot collaboration. The commercial viability lies in its adaptability to diverse assembly environments requiring minimal calibration, applicable across industries like automotive, electronics, and aerospace.