Skip to content
Erwin Lejeune

Erwin Lejeune

Leading robotics teams from the lab to the field.

role
head of robotics @ sirb.ai
based
abu dhabi, uae
leading
teams since 2021
hands-on
ros 2 · c++ · python · go
standards
from embedded & web

Technical lead across AMR navigation, fleet management and fault management, with robotics products shipped end to end. I build teams from scratch, run performance reviews and roadmaps, review pull requests, and set up the AI workflows that make a whole team faster without lowering the bar.

about

Background

I'm Erwin, a technical lead in robotics: navigation for autonomous mobile robots, fleet management and fault management. I've led the teams building them in the lab and in the field, now at SIRB.AI, before that at TII and Coalescent.

I wear two hats. As an engineer, I ship products end to end and stay in the code through design and pull request reviews. As a manager, I build teams, run performance reviews and roadmaps, and set up AI workflows the whole team can rely on.

2018 — 2019 · Hiventive, Ingeniarius

Close to the metal

built
A virtual STM32 in SystemC/TLM running under QEMU, then small ROS robots that won both maze competitions.

2020 — 2021 · Centrale Nantes, Hiventive

Research, and the web

built
Go services with PostgreSQL and Docker for an electronics logistics platform.
led
Four research projects in multi-agent pathfinding, PDDL planning and real-time ROS, with two papers.

2021 — 2022 · Coalescent, Odense

First robot in a real store

built
The navigation stack of a robot working supermarket aisles: planning, control and localization.
led
Navigation and trolley detection after the CTO left: hiring, training, and turning product asks into sprints.

2022 — 2024 · TII, Abu Dhabi

One robot to many

built
Multi-robot orchestration and swarm behaviors, including bird-inspired flocking published at IROS 2024.
led
The R&D and the partner demos, which brought in USD 3M in contracts.

2024 — 2026 · TII, Abu Dhabi

Leading the team

built
A swarm autonomy rework that cut onboard CPU by 74% and bandwidth by 52%.
led
A team of 9 across autonomy, infrastructure and integration, and demos that led to USD 10M+ in contracts.

2024 — now · Unchained Labs

Starting a company

built
Local-first AI agents on clients' own models and data, and Kymatics, a graph-based agent orchestrator.
led
All of it: vision, clients, setting up the company, building and shipping the SaaS.

2026 — now · SIRB.AI, Abu Dhabi

Head of Robotics

built
The robotics architecture and a self-healing layer: fault recording, recovery agents, mission scoring.
led
The autonomy roadmap, from prototype to the public launch at UMEX 2026.

agentic

Adapting to the agentic era

AI changes how software gets built, robotics included. I treat it as a tool that lets a team ship more without lowering the bar, never as a replacement for the people on it. The real work is the harness around it, so quality holds as AI use grows.

how I roll it out

  1. 01 Boring work first: docs, tests, first-pass reviews.
  2. 02 A person owns every change that ships.
  3. 03 AI use grows only as fast as its checks.
  1. draft by a person or an AI
  2. AI review self-hosted, always
  3. checks tests and lint
  4. grip quiz explain your diff
  5. human review a teammate signs off
  6. ship with a clear owner
Every change takes the same path, whoever or whatever wrote it.
  • led

    A company brain in Slack

    Ask it about the codebase, the org or a management question; it answers from sources each team sets up.

  • led

    Self-hosted code reviews

    Quality and code reviews on every change, built and hosted in-house: cheaper than paid tools, same bar.

  • on the robots

    Agents in the loop, not in charge

    They read the fault timeline and propose the repair; deterministic rules still own the actuators.

experience

Experiences

Where I've worked since 2018, one row per company. Pick a bar to see the role.

SIRB.AI
Unchained
TII
Coalescent
ECN
Hiventive
Ingeniarius

Head of Robotics @ SIRB.AI

Feb 2026 — Present · Abu Dhabi, UAE

  • Joined as founding team: designed the robotics software architecture and set the autonomy roadmap.
  • Built a self-healing autonomy layer: swarm-wide fault and state recording, agents that detect and recover from failures, and ML-based mission scoring before launch.
  • Took the autonomy stack from prototype to public launch at UMEX 2026, unveiled as the Middle East's first AI-powered swarm platform.

what I work with

  • ROS 2
  • C++ / Python / Go / Rust
  • Behavior Trees
  • Task & Path Planning
  • Multi-Agent Planning
  • PDDL
  • Sensor Fusion
  • Motion Control
  • Embedded / Real-time
  • Agents / MCP / RAG
  • Docker / CI / RobotOps
  • Scrum / PI Planning

systems

Built for Companies

Systems I've built at the companies I've worked for. These are sketches of how they work; the code and the clients stay private.

plan → formation transit → Voronoi coverage → triangulation

Plan, fly in formation, cover the zone, find the target.

Switching vehicles only means switching a plugin: every motion algorithm sits behind the same tracker interface.

  • Plan around what the fleet must not touch. Zones and every other source of the world are costmap layers; a global plan is routed through them before anything moves.
  • Fly in formation, links in view. The flock holds its slots behind a virtual leader, a safety barrier underneath keeps every vehicle clear of the boundary, and link strength across the mesh is part of the picture.
  • Cover, then localize. In the task zone the formation breaks into a Voronoi partition and each vehicle sweeps its own cell. When one gets a contact, the others take bearings and triangulate it.

projects

Featured Projects

Things I've built on my own time, all open source: planning toolboxes, robotics tools, and the two orgs where my side work lives.

  • personal · in the loop

    grip

    Pass the whiteboard defense before you ship. A git hook that quizzes you on your own diff before you commit or push, grades the answers against a hidden rubric, and blocks anything below your team's pass mark.

    • Python
    • Git Hooks
    • LLM Grading
    Read the docs
  • APLA-Toolbox · in the field

    PyMAPF

    Multi-agent planning on arbitrary graphs, not just grids: CBS, PIBT, LaCAM and MAPF-LNS, plus multi-agent RL. The solvers stream their search live, so you can watch conflicts get found and resolved.

    • Python
    • Multi-Agent Planning
    • Graph Search
  • APLA-Toolbox · in the field

    PythonPDDL

    A dependency-free Python PDDL framework: a hand-written parser and grounder from STRIPS to durative actions, 14 planners, LM-cut and FF heuristics, and heuristics you can train on your own plans.

    • Python
    • PDDL
    • Heuristic Search
  • personal · on the metal

    roue

    Drop a file, get the file you wanted. An offline file converter for the Linux desktop: Ctrl+Shift+right-click a file, pick a format on the wheel, and the result lands next to the original.

    • Python
    • GTK 4
    • FFmpeg
    • libvips
  • personal · on the metal

    rostree

    A ROS 2 dependency graph is a DAG, not a tree. rostree explores it from the command line, a TUI, or a self-contained HTML file that still opens on a robot with no route out.

    • Python
    • Graph Theory
    • ROS 2
    • TUI
  • personal · in the field

    flybots

    Flight algorithms from scratch: multirotor, fixed-wing and VTOL physics written out in full, 40+ runnable simulations, and a gym for teaching a drone to fly itself.

    • Python
    • Flight Dynamics
    • Reinforcement Learning
  • org · on the metal

    Clanker Labs

    Self-host your home, let the agent run it. Notes, mail, media, passwords and money back on hardware you own, with an agent in front so keeping it running doesn't take a sysadmin.

    • Ansible
    • Docker
    • LangGraph
    • MCP
  • org · in the loop

    Unchained Labs

    AI brains for companies: local-first agentic systems that run on their own models, data and infrastructure. Shown here, Kymatics turns a spoken intent into a queue of build jobs.

    • Rust
    • Python
    • Agentic AI
    • Voice / STT

open-source

Open Source

I've contributed to open source since 2018. Robotics runs on shared code, so I give back where I can: algorithms and fixes for the projects I rely on, and tools of my own for problems I kept running into.

  • personal

    awesome-daily-minigames

    281 games that give you one puzzle a day, Wordle and everything since, across 18 categories, with a searchable site generated from one YAML file.

    • Awesome List
    • YAML
    • Node.js
  • personal

    awesome-github-pages-portfolios

    A curated list of portfolio templates and personal sites hosted on GitHub Pages, each with its repo and live demo.

    • Awesome List
    • GitHub Pages
  • personal · in the field

    jarvis_ros2

    Event, fault and recovery management for ROS 2 robots: a real fault lifecycle, rule-based recovery behind safety gates, an agent that explains incidents with evidence, and a live dashboard.

    • ROS 2
    • Python
    • React
    • LiteLLM
  • personal · in the loop

    BTView

    A visual graph editor for BehaviorTree.CPP trees inside VS Code and Cursor: bidirectional XML sync, tidy layout, and a validation panel that jumps to the offending node.

    • TypeScript
    • Behavior Trees
    • VS Code

all 30 projects, sortable and filterable, in the archive →

notes

Writing & Photos

contact

Open Channel

I'm not looking for a new role, but my inbox is open. A question about multi-agent planning, an open-source idea, or just a hello are all welcome, and I'll reply.

Say hello Abu Dhabi, UAE