Loyd Hook

Certification strategies using run-time safety assurance for part 23 autopilot systems

Part 23 aircraft operation, and in particular general aviation, is relatively unsafe when compared to other common forms of vehicle travel. Currently, there exists technologies that could increase safety statistics for these aircraft; however, the high burden and cost of performing the requisite safety critical certification processes for these systems limits their proliferation. For this reason, many entities, including the Federal Aviation Administration, NASA, and the US Air Force, are considering new options for certification for technologies which will improve aircraft safety. Of particular interest, are low cost autopilot systems for general aviation aircraft, as these systems have the potential to positively and significantly affect safety statistics. This paper proposes new systems and techniques, leveraging run-time verification, for the assurance of general aviation autopilot systems, which would be used to supplement the current certification process and provide a viable path for near-term low-cost implementation. In addition, discussions on preliminary experimentation and building the assurance case for a system, based on these principles, is provided.

Toward run-time assurance in general aviation and unmanned aircraft vehicle autopilots

When compared with most other common methods of travel, travel in general aviation aircraft is relatively unsafe. Fortunately, the most frequent causes of fatal general aviation (GA) mishaps could be significantly reduced with very simple autopilot systems. However, such systems can be prohibitively costly due in large part to the expense of validation and verification required to certify them. The Federal Aviation Administration (FAA), NASA, and the US Air Force have been working to develop alternative certification methods to reduce this cost. In particular, run-time assurance (RTA) methods have recently been gaining momentum as a potential avenue to achieve this goal. This has led researchers from the aforementioned group to propose an RTA system for GA autopilots, which uses the human pilot as the baseline controller and a lesser certified autopilot as the advanced controller. This paper expands on that work by developing a hybrid control model which takes into account the human pilots variable timing and control ability. Simulation results and a discussion on the impact of these findings are also provided.

A reliable system design for nondeterministic adaptive controllers in small UAV autopilots

Despite the tremendous attention Unmanned Aerial Vehicles (UAVs) have received in recent years for applications in transportation, surveillance, agriculture, and search and rescue, as well as their possible enormous economic impact, UAVs are still banned from fully autonomous commercial flights. One of the main reasons for this is the safety of the flight. Traditionally, pilots control the aircraft when complex situations emerge that even advanced autopilots are not able to manage. Artificial Intelligence based methods and Adaptive Controllers have proven themselves to be efficient in scenarios with uncertainties; however, they also introduce another concern: nondeterminism. This research endeavors to find a solution on how such algorithms can be utilized with higher reliability. Our method is based on using an adaptive model to verify the performance of a control parameter - proposed by a nondeterministic adaptive controller or AI-based optimizer - before it is deployed on the physical platform. Furthermore, a backup mechanism is engaged to recover the drone in case of failure. A Neural Network is employed to model the aircraft, and a Genetic Algorithm is utilized to optimize the PID controller of a quadcopter. The initial experimental results from test flights indicate the feasibility of this method.

Accounting for helpful and harmful human reactions in run-time assurance frameworks

Run-Time Assurance is a hybrid controller architecture that bounds the behavior of a nondeterministic or intractably complex primary controller by monitoring vehicle state and switching over to a simpler, more straightforward controller that single-mindedly avoids a given hazard. Multiple layers of Run-Time Assurance protections allow for the benefits of a human contribution to aid in upset prevention while maintaining a safety net of automatic recovery in case the pilots action is harmful, insufficient, or nonexistent. This research is an incremental step toward completing the Stall Prevention and Run Time Assurance (SPaRTA) project at the University of Tulsa.

Technology acceptance model for safety critical autonomous transportation systems

It is becoming increasingly clear that a paradigm shift in the way people travel will be seen in the near future. This is due to the ever increasing scope of technology in our lives and a built up public demand for safer, faster, and more efficient transportation options. It is also becoming clear that greater levels of autonomy will enable this paradigm shift to a large degree. However, due to the fact that this will require control over personal safety to be entrusted to the autonomous system, many physiological factors will play an important role in their acceptance. Unfortunately, available technology acceptance models do not include considerations for safety critical systems such as these. This paper proposes a new model which incorporates these considerations focusing on the psychology of control, acceptance, and trust and the factors that influence use of a safety critical technology. This model has been built using data from a series of surveys, simulations, reliability data, and previous technology acceptance models and has been validated using previous research into the usability of autonomous vehicles. The full model and considerations for the improvement of the model as well as further validation techniques is provided. The work in the University of Tulsa Vehicle Autonomy and Intelligence Lab (VAIL) has begun development and verification of the Safety-Critical Technology Acceptance Model and is progressing with the development of the Electronic Car Learning and Intelligence Program Simulator (ECLIPS). Through the investigation of these issues using ECLIPS and user feedback, VAIL is on track to model the acceptance and develop guidelines for the development and implementation of autonomous systems. VAIL is working to research these questions at a fundamental level and describe the topics in a way that can make sure these technologies are in line with the progression of technology and the future of human involvement with these systems.

Dynamic geo-fence assurance and recovery for nonholonomic autonomous aerial vehicles

Currently, there exists a growing demand for increased levels of autonomy in many types of aerial vehicles. This is a direct consequence of the need for increased safety and efficiency as the proliferation of these vehicles increases and the airspace becomes more densely populated. In regard to vehicle autonomy, one need relates to constraining the vehicle to a planned flight path that has set tolerances based on safety and the dynamics of the aircraft. In practice, in order to satisfy this need, potential obstacles, restricted regions, and nondeterministic factors that could lead the aircraft to violate the tolerances of the planned flight path must be accounted for. This paper proposes a solution to this problem which extends the concept of a “dynamic geo-fence” in order to provide for a way to first assure that the path constraints are met and, if necessary, recover to the flight path when a violation occurs. Our solution consists of two parts: a static 2-D vector field map and a trajectory planner. The construction of the vector map takes into account the dynamics of the aircraft and the required path constraints and allows for the presence of obstacles whose shape can be either convex or concave. Additionally, the solution was formulated assuming nonholonomic constraints for the dynamics of the vehicle. Methods for construction of the vector map are provided in this work. Simulation results show recovery trajectories in the presence of complex obstacles, including concave and cluttered environments. Limitations and possible improvements of this approach are also discussed.

Initial designs for an automatic forced landing system for safer inclusion of small unmanned air vehicles into the national airspace

Small unmanned air vehicles (UAVs) have unique advantages and limitations which will affect their safe inclusion into the national airspace system. In particular, challenges associated with emergency handling in beyond line of sight operations will be especially critical to address. This paper proposes initial designs for an autonomous decision system for UAVs to select emergency landing sites in a vehicle fault scenario. The overall design consists of two main components: pre-planning and realtime optimization. In the pre-planning component, the system uses offline information such as geographical and population data to generate landing loss maps over the operating environment, which can be used to constrain planning of flight routes to satisfy a bound on the expected landing loss under worst-case fault. In the real-time component, onboard sensor data is used to update a probabilistic risk assessment for potential landing areas allowing for refinement of the expected loss calculation and landing site selection at the time of a fault. The mathematical models and computational algorithms constituting these system components are based upon methodologies in optimal control and statistical inference. Simulation results are provided to demonstrate the application of the proposed algorithms in an example of quadrotor emergency landing over a section of UC Berkeley campus.