A Contrastive Federated Semi-Supervised Learning Intrusion Detection Framework for Internet of Robotic Things

TL;DR

This work tackles intrusion detection in the Internet of Robotic Things (IoRT) under privacy constraints and label scarcity by introducing CFedSSL-NID, a federated semi-supervised learning framework augmented with contrastive learning. The approach combines server-side labeled data with client-side unlabeled data, using a lightweight 1D-CNN encoder, random weak/strong data augmentation, latent contrastive learning, and exponential moving average (EMA) updates to fuse self-supervised and supervised signals. Key contributions include the integration of FL, SSL, and CL tailored to IoRT, an EMA-based global model refinement, and extensive NSL-KDD-based experiments showing improved accuracy, F1, and robustness with lower resource requirements. The results indicate practical viability for privacy-preserving, real-time IoRT intrusion detection, with future work focusing on further complexity reductions, secure communication, and real-robot validation.

Abstract

In intelligent industry, autonomous driving and other environments, the Internet of Things (IoT) highly integrated with robotic to form the Internet of Robotic Things (IoRT). However, network intrusion to IoRT can lead to data leakage, service interruption in IoRT and even physical damage by controlling robots or vehicles. This paper proposes a Contrastive Federated Semi-Supervised Learning Network Intrusion Detection framework (CFedSSL-NID) for IoRT intrusion detection and defense, to address the practical scenario of IoRT where robots don't possess labeled data locally and the requirement for data privacy preserving. CFedSSL-NID integrates randomly weak and strong augmentation, latent contrastive learning, and EMA update to integrate supervised signals, thereby enhancing performance and robustness on robots' local unlabeled data. Extensive experiments demonstrate that CFedSSL-NID outperforms existing federated semi-supervised and fully supervised methods on benchmark dataset and has lower resource requirements.

Figures (8)

• Figure 1: In pratical IoRT system, server can get labeled data by hiring experts to manually label, while robot clients do not have sufficient capability to label data.
• Figure 2: Brief summary of challenges in practical IoRT scenarios and solutions in our work.
• Figure 3: Overview of the proposed CFedSSL-NID framework. Each robot client k updates $E_\theta^{k}$ by local unlabeled data using $ContrastiveLoss$ and uploads $E_\theta^{k}$ to server. At server-side, global model $E_\theta^{Glo}$ will be updated by EMA with clients' model aggregation. Then $E_\theta^{Glo}$ will be updated by labeled data using $CrossEntropyLoss$ to add supervised signal.
• Figure 4: Example diagram comparing the numerical features of the original IoRT traffic data with its strongly and weakly augmented versions.
• Figure 5: Visualizations of strong/weak augmentations noises and augmented data and original data.