Advancing RF sensing with generative AI: from synthetic data generation to pose completion and beyond

Radio Frequency (RF) sensing has emerged as a pivotal technology for non-intrusive human perception in various applications. However, the challenge of collecting extensive labeled RF data hampers the scalability and effectiveness of machine learning models in this domain. Our prior work introduced innovative generative AI frameworks - RF-Artificial Intelligence Generated Content using conditional generative adversarial networks and RF-Activity Class Conditional Latent Diffusion Model employing latent diffusion models - to synthesize high-quality RF sensing data across multiple platforms. Building upon this foundation, we explore future directions that leverage generative AI for enhanced 3D human pose estimation and beyond. Specifically, we discuss our recent advances in pose completion using latent diffusion transformers and propose additional research avenues: cross-modal generative models for RF sensing, real-time adaptive generative AI incorporating evolutionary learning for dynamic environments, and addressing security and privacy concerns in intelligent cyber-physical systems. These directions aim to further exploit the capabilities of generative AI to overcome challenges in RF sensing, paving the way for more robust, scalable, and secure applications.