Robotic Monitoring Enhancement with Deep Learning and Conformal Prediction for Indoor Anomaly Detection in Emergency Situations

This work investigates how deep learning and conformal prediction techniques might be combined to improve robotic systems’ anomaly detection performance in indoor emergencies. Our method uses the ROSbot 2R mobile robot platform and the YOLOv5 deep learning model to recognize objects in real-time in a Gazebo-crafted simulated environment. Conformal prediction is a useful tool for evaluating prediction reliability, which is important for essential applications like emergency response. Our system is made to recognize and react to certain abnormalities, such as people lying in odd positions or people exhibiting symptoms of possible medical issues. Thorough testing in a range of simulated interior scenarios, such as home and university corridors, demonstrates the efficacy of our method. In addition to advancing robotic monitoring, this research presents a framework for putting into practice trustworthy emergency detection systems that may eventually be useful in real-world situations.