← Projects

Formal verification of a feedback-controlled drone

A drone can stay inside its flight boundary yet remain stuck at the edge. We compared deductive proofs in Lean 4 with model checking in TLA+/TLC and NuSMV to see whether a controller could guarantee both safe flight and eventual return to the interior.

Read the project report ↓
Project team
Can Afacan
Tianrun (Eric) Yang
Course
CE 356 · Northwestern University

Flight requirements

We modeled the flight volume as an ellipsoid with an inner region and a surrounding buffer. When the drone enters the buffer, the controller applies inward braking. We specified three properties for the closed loop.

  • Containment. The drone stays inside the combined flight volume at every step.
  • Speed. Its speed stays within the prescribed limit at every step.
  • Recovery. Each time the drone enters the buffer, it eventually returns to the inner region.

Verification results

We encoded the controller as a finite-state system for TLC and NuSMV. In Lean 4, we expressed each requirement as a theorem and proved it under the model’s assumptions.

MethodResult for the drone model
TLA+ / TLCExplored 6.89 million reachable states in 75 seconds.
NuSMVRepresented 19.74 billion states symbolically in 14.5 seconds.
Lean 4Proved containment, the speed bound, and eventual recovery.

Each state count refers to that tool’s encoding. The guarantees cover the modeled flight dynamics. Wind and other external disturbances were outside the model.

Project report

Open PDF in a new tab

Download full report (PDF)

Oral presentation

Deductive Verification vs. Model Checking for Physical Safety of a Feedback-Controlled Drone
Tianrun (Eric) Yang and Can Afacan
Northwestern Undergraduate Research & Arts Expo, May 21, 2026.

Presentation slides (PDF)Expo program