Hi,
My name is Sam. I would like to help you with your GRU project. I have several years of deep learning experience and have used RNNs quite a bit.
I propose using multitask learning - using x,y coordinates (two features per time step) to predict the most recent x and y separately. This will be written using the keras library.
I will use proper machine learning methodology by splitting data into train/test sets and getting you a decent approximation of generalization (testset) error. I can deliver this code as either a .py or a jupyter notebook. I will also deliver a trained model in .h5 so that you can use it for your purposes without retraining.
Looking forward to discussing this with you further.
Sincerely,
Sam