Future AAM vehicles are expected to incorporate non-traditional components with high degree of intelligence to maintain the vehicle’s operational safety and performance while dealing with unexpected events and emergencies. The development of a comprehensive self-diagnosis and prognosis approach with cognitive properties will play a vital role in AAM operations as it will increase safety assurance by maintaining mission requirements under operational constraints. The general goal of this ULI project is to overcome long-standing limitations in current knowledge that have limited and/or delayed the integration of AAM systems into the NAS. We are developing a safety assurance multi-functional system with self-learning attributes, that enables reliable supervision, management, and control of AAM vehicles. The system generates safety-related specifications that are then integrated to generate optimized safe trajectories and bounded learning control actions under "learned” operational constraints. In addition, we are developing cognitive performance metrics to constantly monitor the efficiency of the intelligent components within the safety assurance architecture.
Existing autonomy frameworks are primarily built around fixed self-properties, such as self-configuration, self-optimization, and self-awareness, defined by preset mission goals and execution strategies. However, these approaches lack the flexibility to adapt to evolving environments, and current assurance methods assume systems remain unchanged after deployment, limiting their ability to learn or evolve. To address this gap, there is a growing need for a dynamic and adaptive safety assurance architecture that can support learning-enabled systems, enhance real-time decision-making, and ensure safe, reliable integration of AAM vehicles into the NAS.
Current trajectory planning and control frameworks are designed primarily to compensate for uncertainties and disturbances without regard to the system’s status, current state, or operational envelope. This limitation could be critical, especially in scenarios where precise actions under unforeseen system degradation must be guaranteed. As a result, the trajectory or control actions that are effective under normal conditions may not fully align with the required safety constraints under new sets of operational conditions. The integration of advanced onboard trajectory planning and intelligent control reconfiguration within a safety assurance architecture will be beneficial for ensuring that precise and safe adjustments are made with a full understanding of the vehicle’s capabilities and limitations.
Future AAM vehicles will likely be operated fully or partially autonomously and will be equipped with learning elements to enhance operational capabilities. Measuring how a system with cognitive properties executes actions, makes decisions, and learns by extracting the appropriate information from a complex collection of data, is a necessary step in certification and represents a significant design challenge. Addressing this challenge requires developing appropriate tools and metrics to monitor the effectiveness of these systems to evolve, scale, and generalize under a variety of scenarios. Linking learning with operational safety by continuously assessing the impacts of the system’s actions as they evolve, will be essential, not only to enhance system performance and reliability but also to build trust in their deployment in sensitive and dynamic environments.
