Wissam BEJJANI

Leading applied RL: from research, to production, to partnerships

Who am I?

I'm an applied RL lead at InstaDeep. My work sits where research meets reality: I embed with clients, take ownership of a hard problem, and carry it from a research idea to a system running in their environment.

There's a simple question underneath all of it that turns out to be very hard: how do you get a learning agent to make good decisions in the real world, where things are uncertain, constraints are physical, and getting it wrong has a cost? I came to that question through robotics, and I still think like a roboticist.

What I do now?
I lead a team of research engineers building multi-agent reinforcement learning systems. Our clients span transportation, robotics, finance, and energy. The job runs the full width of a deployment: understanding the client's problem, building the AI that solves it, and doing the unglamorous work of making it hold up in production.

The project I'm best known for is train dispatching for Deutsche Bahn: an RL system that reroutes hundreds of trains on a live network in real time. It's one of the largest multi-agent RL projects around, and my job is to take it through the hardest stretch, from a system that works in simulation to one that's ready to run in the real world.

The other half of my role is business development: opening new engagements across industries, working out what a client actually needs, and shaping the proposals and partnerships that get a project started. It's the kind of role people increasingly call forward deployed, sitting on both sides at once, close enough to the code to know what's buildable, and close enough to the client to know what's worth building.

Before InstaDeep: robotics?
My background is in robotics and it still shapes how I work. I did my PhD at the University of Leeds on deep RL and imitation learning for robots, teaching them to retrieve objects from cluttered, hard-to-see spaces. Before that, at DLR, I worked on motion planning for a humanoid service robot; that work was later used by an astronaut on the ISS to operate a robot back on Earth. I also spent time at EPFL on probabilistic models for robot motion. Some of the robotics work got a nice bit of press along the way, from the Daily Mail to Communications of the ACM.

Background

Professional Experience

InstaDeep — Team Lead & Senior Research Engineer, Applied RL 2022 – present · France 🇫🇷
- Lead and mentor engineering teams building decision-making AI across transportation, robotics, finance, and energy.
- Team lead on the Deutsche Bahn train-dispatching project, one of the largest multi-agent RL efforts anywhere, driving it toward real-world deployment.
- Drive business development: opening new client engagements across industries, translating needs into solutions, and shaping proposals and partnerships.

University of Leeds — PhD Researcher, Machine Learning & Robotics 2017 – 2021 · Leeds, UK 🇬🇧
- Led research on self-improving robotic object retrieval; built deep RL and imitation learning (IL) methods for decision-making under uncertainty using convolutional and recurrent neural networks (CNN, LSTM).
- Mentored international teams of grad students and interns from different backgrounds and skill levels
- Restructured the Machine Learning course to align the goals and expectations of the students, administration, and field, earning me the qualification "Associate Fellow of the Higher Education Academy, UK"
- Taught Intelligent Systems & Robotics, AI, Practical Programming, and Intro to Web Technologies; mentored international students and interns; managed robotics lab resources.

References: Dr. Mehmet Dogar and Dr. Matteo Leonetti

EPFL — Research Scientist 2016 · Lausanne, Switzerland 🇨🇭
- Developed probabilistic tools using Hidden semi Markov Model (HsMM) for learning the parameters of sequential Dynamical Systems (DSs)

References: Prof. Aude Billard and Dr. Nadia Barbara

Freelance — Technology Consultant 2016 · Anglet, France 🇫🇷
- Raised a plant nursery's greenhouse yield by 15%: analyzed the operation, found the bottlenecks, and deployed a modular wireless system for remote sensing and control of temperature and humidity.

DLR (German Aerospace Center) — Visiting Researcher 2015 – 2016 · Oberpfaffenhofen, Germany 🇩🇪
- Envisioned and prototyped an approach for planning surface wiping commands by a humanoid robot. My contribution was successfully used by an astronaut aboard the International Space Station (ISS) to operate a ground robot (see The METERON SUPVIS experiments)

Reference: Dr. Daniel Leidner

American University of Beirut — Research Assistant 2013 · Beirut, Lebanon 🇱🇧
- Automated the communication routines of an Unmanned Underwater Vehicle operated via the Iridium satellite network
- Developed an augmented reality Android app for archaeological sites

Reference: Dr. Daniel Asmar

Education

PhD, Machine Learning & Robotics — University of Leeds (2017–2021)

Thesis: "Learning Deep Policies for Physics-Based Robotic Manipulation"
Supervisors: Dr. Mehmet Dogar and Dr. Matteo Leonetti

MSc, Mechatronics, Robotics & Automation — TU Dortmund (2013–2015)

GPA 3.7/4.0, German Grade: 1.4
Thesis at DLR: "Automated Planning of Whole-Body Motions for Everyday Household Chores with a Humanoid Service Robot"
Supervisors: Dr. Daniel Leidner and Dr. Oliver Urbann

BE, Mechanical Engineering — NDU, Lebanon (2008–2013)

GPA 3.5/4.0
Thesis: "Thermo-acoustic refrigeration: prototype design and optimization"
Supervisor: Dr. Marwan Azzi and Prof. Charbel Bou-Mosleh

Teaching Experience

University of Leeds, UK 2017 – 2021

Qualification: Associate Fellow of the Higher Education Academy - UK
Courses:

Skills

Technical: multi-agent and applied RL (RLlib), imitation learning, decision-making under uncertainty, robotics and motion planning, Python, JAX, TensorFlow, PyTorch, Docker, HPC, ROS, OpenCV, MATLAB
Leadership & BD: client-embedded delivery from pilot to production, leading and mentoring teams, winning new client engagements, translating needs into solutions, stakeholder presentations, proposals and partnerships

Honors and Awards

Voted 1st place winner of the Flash Presentation Competition at the PostGrad-Research Showcase Event 2018 - Leeds, UK
"Deutschlandstipendium 2014-2015" scholarship for academic excellence and social commitment - Germany
Team leader of the 1st place winning teams of the American Society of Mechanical Engineers (ASME) 2011 Student Professional Design Competition (SPDC) and 2012 Mechatronics Competition (MC) - Middle East region

Invited Talks

Nov `19: International Conference on Intelligent Robots and Systems - Macau, China
Jul `19: Summer School on Smart Environments - Uni. of Balamand, Lebanon
Oct `18: International Conference on Intelligent Robots and Systems - Madrid, Spain
Apr `17: ASME NDU - Notre Dame University, Lebanon
May `16: Robotics Talk - EPFL, Switzerland

Memberships and Alumni

Enactus Hochschule Team (consulting on entrepreneurial projects for developing nations) - Bochum, Germany
American Society of Mechanical Engineers
Scouts of Lebanon (serving 3 years as Chief Scout)
Offrejoie NGO (volunteer)

In the news

Decisive Agents magazine: Fixing the Hard(ware) Problem
digitale-schiene-deutschland: Artificial Intelligence in the Capacity and Traffic Management System
Dailymail: The crème de la crème of AI: British scientists create robot that can move things out of the way to get the milk out of the fridge
Comunications of the ACM: Worker Robots that Learn from Mistakes
German Aerospace Center: An astronaut in space and a robot on Earth working together
Turkiye.ai: Robotun Dusunme Hizi 10 Kat Artti
Chinese Academic Media: Selected Paper from IROS 2019 on Robot Learning

Event Organisation

Apr `19: 3rd UK Robot Manipulation Workshop (212 participants) - Leeds, UK
Apr `18: 33rd ACM/SIGAPP Symposium on Applied Computing (~500 participants) - Pau, France

Languages

English, French, Arabic, and a bit of German

Page updated

Google Sites

Report abuse