Comparison of methods for table tennis ball prediction

Recent developments in computer science have led machine learning to become
one of the most used tools in artificial intelligence in modern times. It is applied in many
areas of practical life and not limited to academia or engineering. This is due to machine
learning’s flexibility that makes it applied easily in a variety of problems areas, which
can be solved if the data is properly managed.
This thesis focuses on using a machine learning approach to a problem that has been
addressed in publications many times using model predictive control.
The similarities that will allow mathematical modelling to be replaced with a machine
learning approach will be analysed and evaluated and ultimately two approaches will be
implemented.
The problem to be solved is to predict a table tennis ball while uncertainties arise in the
sensing process in an average game of table tennis.
The results of this research are compared with current different approaches to solve the
prediction of the ball. The focus and novelty lie in the improved accuracy of the
predictions using suitable neural networks architectures.