Simulink · Autonomous flight control logic in Simulink

Bridging Computer Vision and Control Theory

For the MathWorks Minidrone Competition, I helped design an autonomous flight controller capable of guiding a virtual quadcopter through a complex track using purely visual data. The goal was to translate algorithmic perception into physical motor commands within a formal physics simulation.

Core Architecture

We developed a three-stage autonomous pipeline inside MATLAB & Simulink:

• Perception (Vision Processing): Processed downward-facing camera feeds using the Computer Vision Toolbox to isolate the track path. Calculated precise centroid and angular offsets relative to the drone's body frame to generate real-time error signals.
• Stateflow Logic (Planning): Governed the high-level intelligence using finite state machines. Built robust transitions between take-off, path-tracking, and landing—ensuring the drone stabilizes its altitude before initiating vision processing.
• Flight Dynamics (Control): Translated visual error signals into physical correction commands. Fine-tuned the PID controllers (Pitch, Roll, and Yaw) to ensure the quadcopter wouldn't over-correct or swing violently during sharp turns.

Key Outcomes & Challenges Solved

The primary bottleneck during simulations was the drone overshooting the track on sharp hairpin corners. We resolved this instability by dampening the derivative gain on the yaw controller, which smoothed the control responses dramatically.

Ultimately, the algorithm successfully navigated the course without losing its path lock, completing the simulation in an efficient 53 seconds and demonstrating a strong application of Embodied AI principles.

---

Developed alongside Team Logos.