A Novel Approach to Error Resilience in Online Reinforcement Learning