Detail-Enhanced Attention Metric Convolutional Network for Anomaly Detection in Mobile Robotic Systems

The reliability and safety of robots in various fields have emerged as critical research focal points. Anomaly detection is pivotal for sustaining the normal operation of mobile robots, enabling the timely identification and resolution of anomaly scenarios such as path deviations and sensor failures. Traditional methods often encounter limitations when confronted with largescale data and complex scenarios. This paper introduces a novel deep learning algorithm, i.e., Detail-Enhanced Attention Metric Convolutional Network (DMCN), designed for anomaly detection in mobile robotic systems. A dual-channel embedding module with shared weights is integrated into the DMCN, which automatically extracts and selects salient features from sensor data. The representation of time-series features is optimized, and the Squeeze-Enhanced Axial (SEA) attention mechanism is incorporated to efficiently capture global information and allocate adaptive attention to critical feature channels. The DMCN's effectiveness is validated through experiments using real sensor data of mobile robots, demonstrating its superiority over other methods in terms of detection accuracy.