BTView
A visual graph editor for BehaviorTree.CPP trees, built into VS Code and Cursor. Bidirectional XML sync, tidy layout, and a validation panel that jumps straight to the offending node.
- TypeScript
- Behavior Trees
- VS Code
Erwin Lejeune
Building the layer that decides what happens next.
- role
- lead architect · robot autonomy
- based
- abu dhabi, uae
- domain
- multi-agent planning, behavior trees
- platform
- agnostic — in-store robots, drones
Leading teams that ship mission planning, task allocation and behavior orchestration for drone fleets — today at SIRB.AI and the Technology Innovation Institute. Before drones, the same planning stack drove a robot through the aisles of a supermarket in Denmark. Eight years, four countries, one layer of the problem.
about
Background
Hello! I'm Erwin. My titles have said "robotics," but what I actually build is one layer: given a goal and a set of constraints, decide what happens next. A planner, a behavior tree, a search over a graph — the part underneath the mission that turns intent into action. I don't particularly care which body it's wired to.
Today I lead autonomy architecture at SIRB.AI and the Technology Innovation Institute — mission planning, task allocation, and behavior orchestration for drone fleets. Before drones, I put the same planning stack into a robot working the aisles of a supermarket in Odense, Denmark, at Coalescent Mobile Robotics — same constraints, same discipline, a chassis with wheels instead of rotors. On the side, at Unchained Labs, I apply the same instincts to software agents: giving an LLM a toolset and a plan is the same problem as giving a robot a mission.
My actual background is planning: multi-agent pathfinding, PDDL search, graph theory, real-time control. If the question is "what should this system do next, given what it knows and what it can't do," that's my domain — whether the actor is a drone, a warehouse robot, or an agent with tool access. When I'm not shipping autonomy stacks I'm playing basketball, watching films, or chasing ranked ladders.
Here are a few technologies I've been working with recently:
- Multi-Agent Planning
- Graph Search / A*
- Behavior Trees
- PDDL
- Task Allocation
- Agentic Orchestration
- ROS 2
- PX4 / MAVLink
- C++ / Python / Go
- MCP / Tool Use
- Motion Control
- Docker / CI
experience
Experiences
Eight years, four countries, one thread running through it: what should this system do next. Select a bar for the detail.
SIRB.AI
TII
Unchained
Coalescent
ECN
Hiventive
Ingeniarius
Lead Robotics Engineer @ SIRB.AI
2026 — Present · Abu Dhabi, UAE
- Lead the autonomy team at a defense-technology startup spun off from the Technology Innovation Institute.
- Agentic mission generation for surveillance, monitoring, and tail-chasing — the same planning problem as before, framed as tool calls for an LLM instead of waypoints for a flight controller.
- Architecture for multi-agent task allocation and mission planning across heterogeneous drone fleets.
Robotics Engineer @ Technology Innovation Institute
2022 — 2024 · Abu Dhabi, UAE
- Built the navigation, perception, and mission-behavior stack for heterogeneous drone fleets — the planning and orchestration layer, not the airframe.
- Established behavior-tree-based mission orchestration as the standard pattern for multi-drone tasking.
Senior Robotics Engineer @ Technology Innovation Institute
2024 — 2026 · Abu Dhabi, UAE
- Owned modular software architecture and lifecycle management across the fleet-autonomy stack.
- Scaled task allocation and mission planning to heterogeneous fleets operating concurrently.
Lead Robotics Engineer @ Technology Innovation Institute
2026 — Present · Abu Dhabi, UAE
- Lead a robotics software team of 9 engineers building the planning and orchestration layer for heterogeneous drone fleets — navigation, perception, task allocation, and mission behavior — plus the DevOps/RobotOps infrastructure that gets it onto real hardware.
- Set technical direction for behavior-tree-based mission orchestration across the fleet-autonomy stack.
Principal Solutions Engineer @ Unchained Labs
2024 — Present · Remote
- Built agentic AI tools for automated documentation PRs (Doxmosis), agentic orchestration (Kymatics), contribution analysis, real-estate analysis, and UAE defense data analysis — the planning discipline applied to software agents instead of robots.
- Developed production backend services for LoopFi's blockchain data analysis on EVM chains.
Founding Robotics Engineer @ Coalescent Mobile Robotics
2021 — 2022 · Odense, Denmark
- Founding engineer on the team that delivered the first 24/7 in-store robot at Bilka, Odense — proof the planning stack isn't drone-specific.
- Led the navigation and path-planning stack — localization, motion control, collision-free routing among shoppers — the same planning problem as a drone mission, worked at walking speed instead of flight speed.
- Demonstrations under this stack supported the company's pre-seed round.
Robotics Researcher @ Ecole Centrale de Nantes
2020 — 2021 · Nantes, France
- Led 4 research projects: multi-agent pathfinding, AI planning (PDDL), real-time ROS control on Xenomai/XDDP, and ROS 2 latency benchmarking.
- Published research on real-time jitter measurements under ROS 2 and a survey of multi-agent pathfinding solutions.
Embedded Software Engineer @ Hiventive
2018 — 2019 · Bordeaux, France
- Developed a SystemC/TLM virtual prototype of an STM32 microcontroller for SoC emulation — TIM, I2C, CAN, GPIO, USART peripherals with QEMU integration.
- Built a virtual temperature/humidity sensor and validated firmware against real hardware.
Software Engineer @ Hiventive
2020 — 2021 · Remote
- Built core backend services in Go with PostgreSQL and Docker for a computer-aided production engineering platform.
- Developed an automated programming test manager for candidate evaluation.
Robotics Assistant @ Ingeniarius
2019 · Coimbra, Portugal
- Built full autonomy stacks from scratch with ROS and embedded systems.
- Won both the mapped and unmapped maze-solving competitions.
projects
Selected Builds
Three altitudes, one question: given a goal and constraints, what happens next. A route through a warehouse, a mission for a drone fleet, a sequence of tool calls for an agent — same discipline, different actors. Six with something to show; the rest are in the archive.
-
in the field
PyMAPF
A multi-agent planning toolbox — CBS, PIBT, LaCAM and Prioritized Planning solvers, running on arbitrary graphs, not just grids. Every solver streams its search live, so you can watch conflicts get found and resolved node by node instead of taking the answer on faith.
- Python
- Multi-Agent Planning
- Graph Search
- CBS / PIBT
-
in the field
jupyddl
A pure-Python PDDL planning framework — hand-written parser and grounder covering STRIPS through durative actions, 14 planners from BFS to weighted A* to LM-cut, and heuristics you can train on your own solved plans.
- Python
- PDDL
- A* / Search
- Heuristics
-
on the metal
rostree
A ROS 2 dependency graph is a DAG, not a tree — expanding it path by path is exponential. rostree explores it properly instead: from the command line, a TUI, or a self-contained HTML file with no CDN and no network calls, so it survives being opened on a robot with no route out.
- Python
- Graph Theory
- ROS 2
- TUI
-
in the field
autonomous-uav-guide
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. The platform-agnostic argument made literal — the planning and control layer doesn't know or care what it's flying.
- Python
- Flight Dynamics
- Reinforcement Learning
-
in the loop
Kymatics
Speak an intent and watch it become a queue of build jobs. Voice goes in through a Python speech service, a Rust orchestrator plans and schedules the work, and a React board tracks every job through todo → running → done. The queue plans itself — the same scheduling problem as a robot fleet, with a microphone as the input device.
- Rust
- Python
- React
- Voice / STT
-
in the loop
weave
A live D&D session assistant — captures table audio, transcribes the narrative, maintains a rolling recap, and offers suggestions grounded in the campaign's own characters and lore. STT plus a LangGraph agent plus a memory of the campaign — a planning problem wearing a dice-game costume.
- Python
- FastAPI
- LangGraph
- STT
Other noteworthy projects
-
in the loop -
on the metal setup
One curl command turns a bare Ubuntu box into my entire working environment — shell, toolchains, containers, dotfiles via chezmoi. Ansible underneath for idempotency, a typed Python CLI so every generated command is auditable before it runs.
- Ansible
- Python
- chezmoi
-
in the loop t212-mcp
An MCP server giving AI assistants read-only access to a Trading 212 account — balances, positions, dividends, pies. No code path issues anything but a GET.
- TypeScript
- Node.js
- MCP
-
in the loop Doxmosis
Agentic tooling that keeps documentation alive: watches a codebase, detects drift, and opens documentation pull requests on its own.
- Go
- Agentic AI
- GitHub Apps
-
in the field Bird-Inspired Flocking
Published research on decentralized, acceleration-based coordination for UAVs — a third-order control law for collective motion, validated in field experiments.
- C++
- ROS 2
- Control Theory
- IROS 2024
-
in the loop zucman
A campaign kit for one statistic: sourced research, regenerable charts, decks in two languages, and a zero-dependency dataviz site — all rebuilt from the underlying data by script, never by hand.
- Python
- Matplotlib
- Data Viz
contact
Open Channel
I'm not actively looking, but the inbox stays open — a question about multi-agent planning, an open-source idea, or just to say hi. I'll get back to you.
Say hello Abu Dhabi, UAE