Evolving Connectionist Systems for Adaptive Sport Coaching

Abstract

Contemporary computer assisted coaching software operates either on a particular sub-space of the wider problem or requires expert(s) to operate and provide explanations and recommendations. This paper introduces a novel motion data processing methodology oriented to the provision of future generation sports coaching software. The main focus of investigation is the development of techniques that facilitate processing automation, incremental learning from initially small data sets, and robustness of architecture with a degree of interpretation on individual sport performers’ motion techniques. Findings from a case study using tennis motion data verify the prospect of building similar models and architectures for other sports or entertainment areas in which the aims are to improve human motion efficacy and to prevent injury. A central feature is the decoupling of the high-level analytical architecture from the low-level processing of motion data acquisition hardware, meaning that the system will continue to work with future motion acquisition devices.