In pursuit of autonomous aviation systems that can safely and efficiently operate within the National Airspace, this project will develop and demonstrate a framework for providing algorithmic assurances and designing fault detection, isolation, and recovery (FDIR) methods for those components of the autonomy stack that rely on data-driven methods based on machine learning. In the future, autonomous aviation systems, in the form of unmanned aircraft systems (UAS) and urban air mobility (UAM) services, are expected to result in more than 2.5 million flights per day. These systems will operate at increased levels of autonomy and will extensively leverage non-traditional software components based on machine learning techniques, in tasks as diverse as visual perception (e.g., to detect other aircraft), intent prediction (e.g., to predict future behavior of other agents), and decision-making and control. Such a proliferation of learning-enabled components (LECs) will be driven by their potential to outperform their traditional, non-learning based counterparts (e.g., for object detection and recognition) and enable entirely new capabilities (e.g., fast decision-making in non-stationary environments). However, LECs can be notoriously brittle in practice (as generalization beyond training data is still a poorly understood property) and largely lack appropriate methods for verification and validation. This tendency of theoretical statistical accuracy but demonstrated fragility in practice represents a key barrier for wider and trusted adoption of LECs. In this context, this project has three objectives:
Objective 1: Assurances for Autonomous Systems with LECs: Develop and demonstrate tools and methods to provide assurances for those components within the autonomy stack that rely on machine learning techniques and other similar data-driven techniques.
Objective 2: Run-Time Fault Detection, Isolation, and Recovery for LECs: Develop and demonstrate tools and methods to detect faulty operation for LEC-based autonomous aviation systems, and devise new fault isolation/recovery methods for these systems.
Objective 3: Airspace Management with LEC-based Autonomous Systems: Develop and demonstrate tools and methods to extend the vehicle-centric assurances and FDIR capabilities devised in Objectives 1 and 2 to the airspace system level.
By addressing Objectives 1–3, this project will place algorithmic assurances and FDIR techniques for LEC-based aviation systems on a firm theoretical and algorithmic foundation. This will be pivotal to creating assurance cases for future aviation systems and thus to ensuring the deployment of UAS and UAM systems on a massive scale.
More »Organizations Performing Work | Role | Type | Location |
---|---|---|---|
Stanford University (Stanford) | Lead Organization | Academia | Stanford, California |
Georgia Institute of Technology-Main Campus (GA Tech) | Supporting Organization | Academia | Atlanta, Georgia |
Hampton University | Supporting Organization |
Academia
Historically Black Colleges and Universities (HBCU)
|
Hampton, Virginia |
Massachusetts Institute of Technology (MIT) | Supporting Organization | Academia | Cambridge, Massachusetts |
Massachusetts Institute of Technology Lincoln Laboratory (MIT-LL) | Supporting Organization | FFRDC/UARC | Lexington, Massachusetts |
Raytheon Company | Supporting Organization | Industry | Arlington, Virginia |
University of California-Berkeley (Berkeley) | Supporting Organization | Academia | Berkeley, California |
University of New Mexico-Main Campus | Supporting Organization |
Academia
Hispanic Serving Institutions (HSI)
|
Albuquerque, New Mexico |