Skip to main content Skip to footer Skip to header area

Center for Intelligent Team Robotics & Human-Robot Collaboration

Robot Perception & IoT

Multi-Modal Wireless-Based Cooperative Sensing And Learning Framework For Mobile Robots

Principal Investigator: Kai-Ten Feng

Background

Camera-based monitoring systems pose significant privacy risks and are ill-suited for healthcare settings, while manual supervision is both labor-intensive and error-prone. Existing sensing models rely heavily on manual data labeling, which is costly, time-consuming, and difficult to scale. There is therefore a pressing need for an automated, privacy-preserving indoor human state sensing solution.

 

Research Objectives

This project aims to develop a synergistic system combining Wi-Fi CSI sensing and mobile robots to achieve privacy-preserving human state detection without cameras. The system enables fully automated data labeling via robots (zero manual effort) and employs a hierarchical federated learning architecture to maintain high detection accuracy in dynamic environments.

 

Methods

The project leverages Wi-Fi CSI signals as the sensing modality, with mobile robots performing automated data collection and labeling to train a three-state human detection model (empty, occupied, abnormal). A federated learning mechanism enables distributed multi-node training, complemented by a dual-layer safety mechanism where Wi-Fi detects anomalies and the robot performs precise on-site verification.

 

Innovation

The key innovation lies in replacing manual data labeling with mobile robots, enabling a fully automated sensing model training pipeline with zero human effort. The integration of a dual-layer sensing architecture (Wi-Fi CSI + robot) and a privacy-preserving federated learning framework further provides a scalable, non-intrusive solution tailored for smart healthcare environments.

 

Expected Outcomes

The project expects to deliver a functional Wi-Fi CSI-based human anomaly detection system, with the first phase of robot-assisted detection validated. In the long term, the system will scale into a multi-node federated learning architecture, improving model generalization across diverse indoor environments and providing a privacy-preserving, low-cost intelligent monitoring solution for healthcare settings.