AutoCI
Two LLM agents design a complete imaging system for a task, sensing modality through reconstruction, and every design is validated on scenes it never saw.
95.6% grasping, 90.8% localization
Read more, with 3-D comparison
Ahmed Boubakry
Robotics and computer-vision engineer
M.Sc. in Robotics from EPFL with a minor in Biomedical Technologies, currently researching perception for embodied agents at the MIT Camera Culture Lab. I work on detection, segmentation and sensor fusion for real-world autonomy, in PyTorch, C++ and ROS 2.
MIT Camera Culture LabVisiting student, Cambridge MA
EPFLM.Sc., Lausanne, Switzerland
Selected work
Eight projects across computational imaging, medical image analysis, control and mobile robotics.
Road segmentation from drone video
Segment Anything 2, fine-tuned to segment roads under vehicle occlusion across a full aerial video.
83.2% detection in obstructed scenes
Read more
Synthetic stroke lesion segmentation
Simulated stroke lesions on healthy brain MRI to train a segmentation network without scarce expert annotation.
0.85 Dice, from a 0.73 baseline
Read more
Rover sensor fusion
Fused RGB and lidar for object detection on a Mars rover, sharpening accuracy up close and extending range further out.
Top-10, European Rover Challenge
Read more
Evolutionary robotics
Evolved a modular robot's shape and controller together in simulation, then built the winner in hardware.
Read more
Rocket model-predictive control
Constrained MPC for a thrust-vectored rocket, up to a nonlinear controller that stays stable under actuator delay.
Read more
Autonomous mobile robot navigation
Overhead-camera vision, visibility-graph planning and a Kalman filter that keeps localising when the camera view is lost.
Read more
Card detection and classification
A YOLO detector trained on a hand-annotated dataset to recognise playing cards in live games.
Read more
Experience and research
Research and industry positions in computer vision, perception and robotics, from imaging-system design at MIT to autonomous inspection in industry and medical image analysis at EPFL.
2026, until Sept.
Visiting Student
MIT Camera Culture Lab, Cambridge MA- Built AutoCI, an agentic system that designs task-specific imaging systems end to end, choosing sensing modality, optics and perception algorithm through a propose-and-validate loop between two LLM agents.
- Cut search cost by executing each proposed pipeline rather than retraining one per candidate.
- 95.6% on robotic grasping and 90.8% on localization in cluttered scenes.
2025, 6 mos
Computer Vision Engineer, Intern
Michelin Recherche et Technique, Givisiez, Switzerland- Computer vision for a robot that navigates to vehicles and inspects their wheels: wheel detection, defect detection and weight estimation from onboard sensing.
2025 to 2026, 4 mos
Stroke Lesion Segmentation Semester Project
LTS5, EPFL Signal Processing Laboratory 5, Lausanne- Synthetic lesion simulation and deep-learning segmentation on Human Connectome Project MRI. Raised accuracy by around 10% by augmenting scarce annotations. Graded 5.75/6.
2024 to 2025, 4 mos
Computer Vision Semester Project
LUTS, EPFL Urban Transport Systems Lab, Lausanne- Segmented roads and occlusions from drone video with Segment Anything 2, reaching 83.2% detection accuracy in obstructed scenes. Graded 6/6.
2023 to 2024, 8 mos
Rover Navigation
EPFL Xplore, Lausanne- LiDAR and depth-camera fusion that extended detection range and improved localization, contributing to a top-10 finish at the European Rover Challenge.
2021 to 2025
Teaching Assistant
EPFL, Lausanne- Linear Algebra, Analysis I and II and Advanced Analysis: ran exercise sessions, graded coursework and exams, and proctored examinations. Evening tutor at the Centre Propedeutique.
- Basics of Mobile Robotics (MICRO-452), 2024: graded student projects and exercises and supported teams through the term.
Skills
Tools and methods used in the work above, grouped by what they are for.
Languages
PythonC++MATLAB
Vision & deep learning
PyTorchOpenCVMONAIYOLOSegment Anything 2scikit-image
Robotics
ROS 2GazeboPath planningMotion control
Sensing & fusion
LiDARStereo & depth camerasIMUKalman filteringCamera calibration
Control & estimation
Model-predictive controlCasADiYALMIPLQRState estimation
Simulation & 3-D
Mitsuba 3Open3DPoint cloudsMesh reconstruction
Medical imaging
NiBabelSynthSeg3-D U-NetNIfTI / DWI & ADC
Tooling
GitDockerLinuxWeights & Biases
Away from the screen
I travel to see how other people live. So far France, Morocco, Spain, South Korea, the United States and Japan, the last of these twice: once with my section at EPFL and once with friends.
Football is a hobby of mine and I play regularly. Team sport keeps up the habits that matter just as much on an engineering team: communicating quickly under pressure, covering for someone when a plan falls apart, and treating the result as collective rather than individual. I also play video games.
France
Morocco
Spain
South Korea
United States
Japan
