Robust Multi-Objective Preference Alignment with Online DPO