SuperBots - Robotic Context Inference Utilizing Computer Vision and Through-Wall WiFi Sensing

SuperBots is a multi-modal autonomous robotic system designed to perform context inference in unknown indoor environments through the fusion of computer vision and WiFi-based through-wall sensing. This proposal describes the WiFi sensing subsystem, which enables the robot to detect, identify, and classify human activity without requiring line-of-sight, even through physical barriers. Deployed on a TurtleBot 4 platform equipped with an Intel 5300 WiFi card and an NVIDIA Jetson Orin Nano, the system captures Channel State Information (CSI) from propagating radio frequency signals whose multipath reflections encode distinctive signatures of human motion and gait. Raw CSI data is processed through a pipeline incorporating noise filtering, principal component analysis, and time-frequency spectrogram generation, with extracted features fed into machine learning models for person identification and activity classification. The system is evaluated under both line-of-sight and non-line-of-sight conditions across varied indoor environments. A key innovation is the use of a video-to-RF simulation framework, which addresses the data scarcity challenge inherent to RF sensing by generating synthetic training data from real video footage, improving model generalizability without requiring extensive physical data collection. The WiFi subsystem is designed for seamless integration with the project's navigation and computer vision pipelines via ROS 2, enabling cross-modal validation that compensates for the limitations of each individual sensing modality. Together, these subsystems advance toward a robust, infrastructure-free platform for through-wall human perception in security, healthcare, and emergency response applications.