Theo Guegan
I’m a robotics and machine learning engineer working on VLA for robotic control.
Experience
Working on physical agents
- Architected a real-time embedded Lua scripting engine in modern C++ (TDD) for on-drone customization, reducing mission prototyping time by 4×.
- Streamlined build processes using Makefiles and an Alchemy build system for cross-compilation and deployment to embedded Linux targets.
- Integrated a local LLM using Rust and Docker for natural-language drone commands, achieving 85% accuracy and demonstrating advanced system integration.
- Contributed to hardware/software debugging and validated system behavior in real-world scenarios, including a high-profile live demo.
- Designed and implemented target-based navigation, adaptive cruise control, and obstacle avoidance algorithms in MATLAB, achieving 99% safety in simulation.
- Led system integration, porting control software to a Renault Zoe using Python and ROS for on-hardware validation and testing.
- Directed the development of an autonomous navigation stack that secured 1st School Award (2024) and Open Category (2025) 🥇.
Led technical project execution for clients like Airbus, signing over €30,000 in projects by coordinating engineering students.
Education
Deep Learning, Robotics. GPA 4/4
Embedded Computing & Autonomous Systems. GPA 5/5
Projects
Collaborated with a multidisciplinary team to design our vision of future space exploration using the HeRo 2.0 swarm robotics platform. Over three days, we worked hands-on from PCB soldering and hardware assembly to high-level autonomous coordination software, integrating perception, communication, and swarm behavior. Our project earned 2nd place among 100 participants for its technical depth and cohesive vision of scalable robotic exploration.
Designed and implemented a hybrid CNN-Transform with PyTorch to decode 512-channel neural spike trains into text. Built the training & evaluation pipeline with custom dataloaders, batching, and CTC loss.
Developed neural surrogate models for Model-Predictive-Control (MPC) behavioral cloning. Benchmarked with both offline and online metrics inside a MuJoCo simulation environment, achieving 90% accuracy.
"Aide-un-étudiant" is a first-place-winning hackathon project. A student mutual aid platform fostering local solidarity through object lending, service exchange, and knowledge sharing. Designed with responsible digital principles, it prioritizes accessibility, performance, and sustainability, including a Positive Impact Score to encourage eco-friendly actions.
Designed a distributed, cross-platform payment application in Rust. Implemented peer-to-peer communication inspired by "Pay'UTC" without relying on central servers. Full-stack rust implementation with Dioxus
Designed a multi-drone coordination system enabling autonomous following behavior. Supported MAVSDK and Parrot Olympe protocols for cross-platform deployment.