SportsNGEN: Generating Realistic Sports Gameplay with Machine Learning
Core Concepts
SportsNGEN is a transformer decoder model trained on sports player and ball tracking data, capable of generating sustained and customizable gameplay simulations for tennis and football.
Abstract
Introduction: Discusses the benefits of machine learning in sports applications.
Trajectory Prediction: Challenges in predicting player and ball trajectories over time.
Reinforcement Learning: Utilizing RL techniques for sports simulation.
Methodology: Describes the approach to generating sports simulations using SportsNGEN.
Tennis Implementation: Details the implementation of SportsNGEN for tennis matches.
Experiments: Evaluates the system's performance in simulating tennis matches.
Impact Statement: Highlights the potential societal consequences of the research.