Reinforcement Learning-Based Model Matching to Reduce the Sim-Real Gap in COBRA