I build computer vision systems for robots, from camera calibration and 3D reconstruction
to model training and deployment. At Preen, I helped take an autonomous robotic car wash from
prototype to operation at a customer site. Previously, I worked on onboard perception for
nano-drones at IDSIA.
Technical skills
Computer vision & 3D
Multi-camera calibration, RGB-D reconstruction, pose and displacement estimation,
image segmentation, photogrammetry, OpenCV, Open3D;
exploratory work with NeRF and 3D Gaussian Splatting
Machine learning
PyTorch, CNNs, synthetic data generation, model evaluation, quantisation,
on-device inference
Python, C (working knowledge), Docker Compose, Linux, AWS, Git, CI/CD, networking,
Raspberry Pi fleet provisioning
Languages
Italian (native), English (full professional proficiency)
Experience
Computer Vision & Machine Learning Engineer
Preen Technologies AG, Zurich· to present
Calibrated a 40-camera RGB-D rig and developed its 3D reconstruction pipeline.
Validated reconstructed geometry against physical vehicles to millimetre agreement.
Trained and deployed a Co-DETR segmentation model for car-part identification and
dirt estimation. Rebuilt the evaluation around a representative test set and
reproducible training, confirming the improvement in customer-site comparisons.
Built a Blender pipeline that renders CAD car models into photorealistic training
images, with segmentation labels generated from part properties. The images later
became part of the Co-DETR training set.
Built a simulation to optimise camera count and placement around vehicles under
coverage, spacing and symmetry constraints. Compared genetic algorithms and grid search,
and validated simulated camera outputs against the production data format.
Designed and benchmarked a silhouette-to-mesh method to estimate vehicle displacement
and yaw from existing cameras. Tests against physically measured ground truth exposed
parallax limits in the camera arrangement and informed the move to ultrasonic sensing.
Built the ultrasonic correction system that went into production: fused four
distance measurements to estimate changes in vehicle position and yaw between scanning
and washing, then applied the correction to the robot trajectories.
Developed and maintained robot path generation across different vehicle geometries,
fixed production defects, and extended regression tests to the full production dataset.
Deployed the service stack with Docker Compose across GPU generations. Separately
provisioned a 40-node Raspberry Pi camera fleet, with custom-built live camera health
monitoring and remote testing and debugging tools.
Machine Learning & Embedded Systems Engineer
IDSIA USI-SUPSI, Lugano· to
Developed CNNs to estimate a peer drone's relative pose and LED state from
low-resolution onboard images, covering data collection, training and flight testing.
Compared 17 CNN architectures, quantised the selected model to 8-bit, and deployed it
on a GAP8 microcontroller at 48.3 frames per second and 95.4 mW. Validated autonomous
tracking for up to two minutes of continuous flight.
Computer Vision Engineer, Digital Pathology
IDSIA USI-SUPSI, Lugano· to
Built a U-Net pipeline to detect malignant cells in bright-field and fluorescence
microscopy. Prepared datasets, handled class imbalance through augmentation and
sampling, evaluated precision and recall, and delivered a standalone prediction
script to the client.
Data Scientist & Software Engineer
Enet Energy SA, Lugano· to
Evaluated an LSTM trading strategy and showed it was unprofitable in paper trading,
informing the decision not to deploy it live. Ported it from MATLAB to Python and
built a real-time MACD system with a custom grid-search optimiser.
Publications
Ultra-low Power Deep Learning-based Monocular Relative Localization Onboard
Nano-quadrotors. S. Bonato, S. C. Lambertenghi, E. Cereda, A. Giusti, D. Palossi. IEEE ICRA , pp. 3411–3417.arXivIEEE
Vision-state Fusion: Improving Deep Neural Networks for Autonomous Robotics.
E. Cereda, S. Bonato, M. Nava, A. Giusti, D. Palossi. Journal of Intelligent & Robotic Systems, vol. 110, article 58, .arXivSpringer
Cyber Security aboard Micro Aerial Vehicles: An OpenTitan-based Visual
Communication Use Case.
M. Ciani, S. Bonato, R. Psiakis, et al. IEEE ISCAS .arXivIEEE
Professional service: Reviewer, IEEE/RSJ IROS .
Education
MSc, Artificial Intelligence
USI Università della Svizzera italiana, Lugano· to
