Navya M Gowda You could use a feedforward neural network with fully connected (dense) layers. The input layer should take the commanded position (x, y) as inputs, and the output layer should produce the estimated joint angles. E.g.:

def create_model():

model = tf.keras.Sequential([tf.keras.layers.Dense(64, activation='relu', input_shape=(2,)), tf.keras.layers.Dense(64, activation='relu'), tf.keras.layers.Dense(2) # Output layer with 2 neurons for joint angles])

return model

Christian Schmidt Thank You