Value Functions for RL-Based Behavior Transfer: A Comparative Study