Using input layer before using LSTM layer is causing issues

Question

I am trying to figure out a way to create a custom model that accepts a sequence with 300 timesteps with 3 features (x, y, touch). Consequently, I used Keras\'s input layer