Interactive RF Game Design for Deciphering Real-World Human Motion: Activities, Gestures, and Sign Language

Current methods for acquiring datasets for human motion classification are limited to controlled settings where participants are directed by a human experiment organizer. Datasets acquired in controlled settings often cannot capture natural human behaviors and are inadequate for obtaining large amounts of real-world data in a sustainable fashion. This paper proposes a new paradigm for automated acquisition of natural human movements based on Interactive RF Gaming. The training of AI/ML models is considered over an evolution of time: pre-deployment via physics-aware batch training and post-deployment via continual learning from interactions. Algorithms needed to address real-world considerations such as the parsing of continuous streams of data, computational constraints, real-time processing, and game design considerations based on cyber-physical human system requirements are also discussed.