Abstract: Vertical take-off and landing (VTOL) unmanned aerial vehicles (UAVs) are versatile platforms widely used in applications such as surveillance, search and rescue, and urban air mobility. Despite their potential, the critical phases of take-off and landing in uncertain and dynamic environments pose significant safety challenges due to environmental uncertainties, sensor noise, and system-level interactions. This paper presents an integrated approach combining vision-based sensor fusion with System-Theoretic Process Analysis (STPA) to enhance the safety and robustness of VTOL UAV operations during take-off and landing. By incorporating fiducial markers, such as AprilTags, into the control architecture, and performing comprehensive hazard analysis, we identify unsafe control actions and propose mitigation strategies. Key contributions include developing the control structure with vision system capable of identifying a fiducial marker, multirotor controller and corresponding unsafe control actions and mitigation strategies. The proposed solution is expected to improve the reliability and safety of VTOL UAV operations, paving the way for resilient autonomous systems.
Introduction
Autonomous Vertical Take-Off and Landing (VTOL) aircraft, such as tiltrotors, represent the "Swiss Army knives" of modern aviation, combining the hovering capabilities of helicopters with the efficient cruise of fixed-wing aircraft. However, the transitions between these states—specifically during take-off and landing—are the most precarious phases of flight. Traditional safety analysis often focuses on component reliability (e.g., "What happens if a motor fails?"). Yet, in complex autonomous systems, accidents frequently occur when all components function correctly but interact in unforeseen, unsafe ways.
This research, a collaboration between UIUC and MIT, addresses the lack of a holistic safety framework for these systems. By integrating vision-based flight control with System-Theoretic Process Analysis (STPA), the authors identify hidden risks arising from the complex interactions between vision algorithms, state estimators, and flight controllers.
System-Theoretic Process Analysis (STPA)
Unlike traditional reliability engineering, STPA treats safety as a dynamic control problem rather than a component failure problem. The researchers developed a multi-level hierarchical control structure to identify Unsafe Control Actions (UCAs). This architecture spans from the remote pilot and high-level autopilot down to individual sensor processing units.
By modeling the vision system—specifically using AprilTags (barcode-like fiducial markers)—within this control loop, the authors could pinpoint exactly where a "correct" vision signal might lead to an unsafe aircraft response due to timing, feedback delays, or environmental interference.
Key Insights and Safety Mitigations
A central finding of the study is that safety is fundamentally an interaction problem. For instance, a vision system does not just fail by "not seeing"; it can fail by providing data "too late" (UCA-8), which is just as hazardous as a total sensor failure.
Another critical discovery involved the "Altitude Trigger." Many vision-related hazards are exacerbated by incorrect altitude estimates. If the system incorrectly believes it is at a lower altitude than it truly is, it may stop searching for landing markers prematurely, leading to a loss of control. To mitigate these risks, the authors proposed several technical solutions:
- Multi-Marker Fusion: Utilizing multiple AprilTags of varying sizes ensures that as the UAV descends, at least one marker remains within the camera's optimal field of view and resolution.
- $L_1$ Adaptive Control: This is employed to handle model mismatches and external disturbances like sudden wind gusts, ensuring the aircraft adheres to its reference trajectory even under uncertainty.
- Secondary Sensing: Integrating infrared (IR) sensors allows the system to cross-verify altitude data, providing a fail-safe against vision-only estimation errors.
Conclusion
The paper provides a rigorous taxonomy of eight specific UCAs for VTOL landing, categorizing scenarios where logic errors or environmental occlusions lead to unsafe states. While the current analysis focuses on the hover phases (take-off and landing), it establishes a robust framework for future work. The next steps involve experimental validation on physical prototypes to quantify the improvement in safety metrics and extending the STPA analysis to the complex transition phase between vertical and horizontal flight.
Summary updated