SuperBots: Context Inferencing using Computer Vision and WiFi based through wall sensing

SuperBots is a multi-modal autonomous robotic system designed to perform real-time context inference in unknown environments through the fusion of computer vision and WiFi-based through-wall sensing. Deployed on a TurtleBot 4 platform equipped with an NVIDIA Jetson AGX Orin and an OAK-D Lite camera, the system integrates object detection, facial recognition, activity classification, and probabilistic localization to enable semantic understanding of human presence and behavior. The computer vision pipeline leverages state-of-the-art deep learning models, including YOLOv4, SSD, FaceNet, and lightweight pose estimators, optimized via TensorRT and ONNX quantization to meet a target inference latency of under 100 ms per frame. WiFi Channel State Information (CSI) is used to extract gait features that complement visual detections in occluded or non-line-of-sight conditions, with a target fusion recognition accuracy exceeding 85%. Autonomous navigation is achieved through ROS 2 Nav2 integration using Smoothed A*, DWA local planning, and AMCL-based localization. Target applications include search and rescue in disaster scenarios, fall detection in healthcare settings, and general-purpose autonomous perception in GPS-denied, unstructured environments. The system aims to push the frontier of on-device, multi-modal perception on compact robotic platforms.