An Empirical Comparison of Off-policy Prediction Learning Algorithms on the Collision Task