Title:	MARS: Multi-Agent Representation learning for Sports data
DNr:	Berzelius-2024-440
Project Type:	LiU Berzelius
Principal Investigator:	Atsuto Maki <atsuto@kth.se>
Affiliation:	Kungliga Tekniska högskolan
Duration:	2024-11-29 – 2025-06-01
Classification:	10207
Keywords:

Abstract

This project seeks to advance automatic interpretation of football tracking data using machine learning and computer vision techniques. Our goal is to develop foundation models for football language that can understand game dynamics, focusing on modeling the complex spatial and temporal correlations of 22 players and the ball on the pitch. At the current stage, we aim to - Develop a representation learning pipeline for football skeleton tracking data We plan to employ self-supervised learning on large graph structures that map the spatial and temporal connections between players' joints and the ball. Given the scale and complexity of this data, substantial computational resources will be required to support effective training. - Establish a skeleton-aware Expected Goal (xG) model. The xG model will estimate the probability of a shot resulting in a goal, representing the quality of scoring opportunities based on player and ball positions. The xG task will not only make the project practically valuable but also serve as a reliable benchmark for evaluating model performance and guiding iterative improvements. This project is related to TRACAB AB and KTH, and partially supported by the Wallenberg AI, Autonomous Systems and Software Program (WASP) funded by the Marianne and Marcus Wallenberg Foundation.