Case Studies

This page presents several of my projects as case studies: for each one, the modelling, the solution and the quality of the results are laid out.

• Drones
• PID control
• Vision
• Delayed systems

Precision landing of a Parrot drone

Land a 30 cm drone in a 60 cm area marked by a QR code, while all the computation runs on a remote PC.

Key resultGood tuning is not enough: delays, noise and packet losses erode stability, and the controller update rate matters most.

Read the study → See the mission (in French)

• Thermal
• PI
• MPC
• Reinforcement learning

Controlling a 450 m heated pipe

Regulate the outlet temperature of a 450 m loop close to boiling point, with a long transport delay.

Key resultPI, MPC and RL compared: with an excellent model, a well-tuned PI meets the needs in the vast majority of cases.

Read the study → See the mission (in French)

• Reinforcement learning
• AIS data
• Anomaly detection

Fishing-boat reinforcement learning environment

Simulate fishing vessels in Dakar's exclusive economic zone to generate synthetic examples of illegal fishing, which are rare in real data.

Key resultAn environment built on real maps and AIS trajectories, where agents learn fishing or evasive behaviours, reusable for anomaly detection.

Read the study → See the mission (in French)