This paper presents a fault detection and diagnosis (FDD) method to enhance the reliability and safety for longitudinal control of an autonomous all-terrain vehicle (ATV). An integrated approach using decentralized and centralized FDD is proposed to optimize the tradeoff between sensitivity and robustness. While the decentralized approach is suitable for detecting faults in actuators and sensors directly connected to a single processor, it is sensitive to noises and disturbances and thus may result in false alarms. On the other hand, the centralized approach is based on information communicated between multiple processors, and it detects and diagnoses faults through analyzing concurrent computations of multiple hardware modules. However, its performance is still limited to isolating faults specifically in terms of components in the single hardware. To incorporate the advantages of both FDD approaches, a two-layered structure integrating both decentralized and centralized FDD is proposed and allows us to perform more robust fault detection as well as more detailed fault isolation. Finally, the proposed method is validated experimentally via field tests of an ATV.