Combining Pre-Trained Models for Enhanced Feature Representation in Reinforcement Learning