Failure Aware Supervision for Low Altitude UAV Swarms

Problem Framing

UAV swarms often show the same outward symptom when things go wrong: tracking error, formation drift, or reduced coordination quality. The harder problem is that those symptoms can be caused by fundamentally different failure modes. A wind disturbance affects the dynamics layer. GPS drift or sensor corruption affects the observation layer. Packet delay or loss affects the communication layer.

A purely reactive controller responds to all of these with essentially the same correction logic. This work asks whether a swarm can do better by keeping a dependable inner controller and adding a lightweight supervisory layer that interprets the failure context before adjusting behavior.

Architecture

The system uses a layered control design. At the bottom, each vehicle keeps a local PID loop responsible for stabilization and reference tracking. Above that, a supervisory layer monitors fault context and applies bounded adjustments to trajectory references, formation scaling, and coordination behavior.

The design is intentionally cross layer. Disturbances are not treated as isolated events inside one subsystem. Instead, the controller considers the interaction between physical disturbances, sensing reliability, and neighbor connectivity so that the swarm can adapt in a way that remains interpretable and operationally realistic.

• Inner loop stabilization stays classical and predictable.
• Supervisory adaptation remains bounded rather than fully free form.
• Coordination behavior changes according to diagnosed fault type instead of generic error alone.
• The framework is designed to fit edge native autonomy rather than cloud only orchestration.

What The Research Actually Builds

• A reproducible simulation environment for multi UAV tracking and coordination under adverse operating conditions.
• Fault injection pathways for wind disturbance, sensor corruption, GPS drift, and communication degradation.
• A diagnosis aware supervisory layer that modifies effective reference behavior without discarding the underlying PID controller.
• A communication aware framing of swarm behavior that connects control decisions to low altitude wireless network conditions.
• A repeated trial evaluation setup designed to study degradation, recovery behavior, and robustness under changing stress windows.

Scenario Design

• Nominal tracking missions used to confirm that supervision does not destabilize the baseline controller.
• Fault injected missions that stress the swarm with changing wind patterns, sensing uncertainty, and network disruption.
• Low altitude wireless network scenarios where communication state becomes part of the control story rather than a separate afterthought.
• Mission inspired cases such as persistent aerial patrol, monitoring, and disturbance prone edge assisted operations.

Why This Matters

A lot of autonomy work either stays purely in controls or purely in networked intelligence. This thesis sits in the overlap. The practical question is how to keep a swarm useful when the air, sensors, and links all stop behaving ideally at the same time.

That is relevant for defense adjacent systems, resilient inspection workflows, emergency response, and low altitude wireless networks where mobility and communication are tightly coupled. The contribution is not just a controller. It is a systems perspective on how bounded decision loops, coordination, and supervision can coexist in a deployable architecture.

Publication Safe Summary

• The public version emphasizes architecture, methodology, and systems framing.
• Detailed benchmark numbers and comparative claims are intentionally withheld.
• The research direction includes both swarm control and agentic low altitude wireless network coordination.
• The work is currently being shaped into publishable material, so this page is meant to communicate depth without disclosing review sensitive content.