Road vehicle safety is of major importance for autonomous vehicles. Complexity deriving from both the system itself and its external environment can lead to failures, i.e., breakdowns in adaptation of coping with such complexity. Resilience engineering provides a distinctive sight for measuring and maintaining safety for complex systems, where safety is seen as something positive from the adaptability view, emphasizing the ability to proactively adjust to hazards and not lose functionality. In the present work, a resilience evaluation framework is proposed for autonomous vehicles. It considers the deviation from a desired performance and its oscillation as two main factors that drive the system/process resilience. Two case studies involving the control module and the perception module of autonomous vehicles are demonstrated under the proposed rationale. In the first case study, a modified pure-pursuit controller that considers lateral error and oscillation information is developed to improve the performance of lateral control. In the second one, a hierarchical neural network is presented, which uses position prediction error as the supervision information to enhance the adaptability of the positioning system in GNSS outage. Simulations that use the Carla simulator and field tests on a real autonomous vehicle platform corroborate the rationale for resilience assessment proposed for operation of autonomous vehicles. Results also show that the metrics used are highly effective in driving recovery operational decisions that restore the functionality of the vehicle in case of disruption, enhancing its resilience, and consequently their safety in operation.