BREAKING — Robot rehearses jet-engine surgery in simulation +++ PPO agent balances pendulum after 2M steps +++ Hamburg's bus stops now planned in VR +++ YOLO model spots cars from above +++ Student claims C++ and Python can coexist peacefully +++ Solar panels installed, planet marginally cooler +++ BREAKING — Robot rehearses jet-engine surgery in simulation +++ PPO agent balances pendulum after 2M steps +++ Hamburg's bus stops now planned in VR +++ YOLO model spots cars from above +++ Student claims C++ and Python can coexist peacefully +++ Solar panels installed, planet marginally cooler +++
Hamburg, GermanyVol. I · No. 1
|
Das Portfolio
“All the Code That’s Fit to Print”
Wednesday, June 10, 2026•dynotic.de•Price: free of charge
The Lead Story
Notes on Making Machines Move
Robotics in simulation, city infrastructure in XR, the occasional solar panel — built quietly in Hamburg.
By ANIRBAN DUTTA — Robotics Desk, Hamburg
From the Editor
The Work Should Speak First
HAMBURG — I study engineering sciences at TUHH and spend most of my time on software that meets the physical world: a policy that moves a joint, a model that reads a drone image, a pipeline that helps keep an engine flying. Most of it is simulation — the patient business of letting things fail a thousand times cheaply before they may fail once, expensively.
This page is set like a newspaper so the work can carry the story; the sections below hold the positions, the projects and the tools, and they say it better than I would. What began with solar panels on school roofs continues with robot arms — the thread is the same: understand a system from the ground up, then build something real. I’m looking for work in robotics simulation.
fig. 1 — A robot arm in twelve shades of punctuation. After a1k0n’s spinning donut.
Experience
Feature stories from the field
HAMBURG — Jun 2026 – present
Teaching Robots to Repair Jet Engines From the Inside
Working Student (Werkstudent), Robotics Team — DLR Institute of Maintenance, Repair and Overhaul (MRO), Hamburg · Project GreenMES
GreenMES is a joint project of the German Aerospace Center’s Institute of Maintenance, Repair and Overhaul, Lufthansa Technik and TUHH’s Institute of Aircraft Production Technology: repairing aircraft-engine compressor blades while they remain installed — no disassembly, no downtime measured in weeks.
Blades deep inside a high-pressure compressor can only be reached through openings a few centimetres wide; that demands miniaturized kinematics, and a great deal of simulation. My part: building and validating the robotic workflows in NVIDIA Isaac Sim and ROS 2 before they ever touch an engine.
Research Assistant (HiWi), MAICoS — Molecular Analysis for Interfacial and Confined Systems · 20 hrs/month
MAICoS is an open-source Python toolkit, developed at TUHH, for analyzing molecular-dynamics simulations of interfacial and confined systems. The work here: new analysis modules and simulation pipelines for materials-science research — public and peer-reviewed.
Python
NumPy
Open Source
MD Simulation
HAMBURG — Jul 2024 – Sep 2025
The City That Plans Its Bus Stops in Virtual Reality
Working Student, XR DigiLab — LSBG (Hamburg’s Agency for Roads, Bridges and Waters) · 20 hrs/week
Fourteen months of open-source Unity and WebXR applications for the City of Hamburg — among them a 3D CAD editor for bus-stop planning on Meta Quest 2 and Apple Vision Pro, developed under the Hamburg-Takt initiative.
On the research side: YOLO models trained on drone imagery to classify vehicles for smart-traffic studies, and Munich’s open-source traffic portal DAVe rebuilt for Hamburg’s digital-twin data layer — Docker, Java, PostgreSQL and Keycloak included.
And an indoor-navigation app in Unity with AR Foundation and Cloud Anchors, guiding visitors through public buildings by SLAM-based spatial anchoring.
Unity
WebXR
YOLO
Docker
Java
PostgreSQL
Keycloak
AR Foundation
SLAM
✂ — below the fold —
Projects
Published work & upcoming editions
PublishedMarch 2026 · Bachelor-thesis groundwork
PPO, From Scratch
PPO-Clip (Schulman et al., 2017) implemented in plain PyTorch for Gymnasium’s Pendulum-v1, without RL frameworks — an exercise in understanding every moving part. The architecture is laid out for 6-DOF joint control of the PAROL6 desktop robot arm, with modular environment wrappers and a sim-to-real transfer plan built on domain randomization.
PyTorch
Gymnasium
Reinforcement Learning
Python
ArchiveAug 2019 – Aug 2021 · Founder
Thirty Students, Four Countries, One Sun
Before the robots, renewable energy: LEAP, a student environmental initiative — thirty students in four countries putting solar panels on school roofs, and €3,000 crowdfunded for lake-restoration projects together with the Rotary Club and the IIT Alumni Association.
Leadership
Crowdfunding
Solar
In the Pipeline — Coming Editions
Four robotics-simulation builds, scheduled for print
PlannedQ3 2026 · ~10 weeks
PAROL6 Grasp Lab — Sim-to-Real Pick & Place
A complete manipulation pipeline for the PAROL6 desktop arm: URDF modeling, motion planning with MoveIt2 on ROS 2, and a reinforcement-learning grasp policy trained in NVIDIA Isaac Lab with domain randomization — benchmarked against the classical planner before a staged sim-to-real transfer. Builds directly on the from-scratch PPO work above.
ROS 2
MoveIt2
Isaac Sim / Lab
URDF
PyTorch
C++ / Python
Editor’s note:Manipulation plus sim-to-real is the single most requested portfolio piece in robotics-simulation interviews — and continuing an existing project shows follow-through.
PlannedQ4 2026 · ~12 weeks
Aisle Nine — Autonomous Warehouse Navigation in Isaac Sim
A differential-drive mobile robot navigating a photorealistic warehouse in NVIDIA Isaac Sim: Nav2 with custom behavior trees, costmap tuning for dynamic obstacles (forklifts, pedestrians), and fused 2D-lidar + depth-camera perception. Ships with reproducible benchmark scenarios — success rate, time-to-goal — and Docker-based CI.
ROS 2
Nav2
Isaac Sim
Behavior Trees
Sensor Fusion
Docker
Editor’s note:Nav2 plus Isaac Sim mirrors the exact production stack of warehouse-robotics teams across the US and EU; metrics and CI make it read like industry work, not a tutorial.
PlannedQ1 2027 · ~8 weeks
FuseKF — Sensor-Fusion Localization in Modern C++
An Extended and an Unscented Kalman filter written from first principles in modern C++ as ROS 2 nodes, fusing wheel odometry, IMU and GPS. Validated in Gazebo and on recorded rosbags, benchmarked head-to-head against the standard robot_localization package — with RViz visualizations and unit-tested math.
C++17
ROS 2
Gazebo
Eigen
robot_localization
GTest
Editor’s note:From-scratch state estimation in C++ answers the two questions every robotics interviewer asks: can you do the math, and can you write production C++?
PlannedQ2 2027 · ~8 weeks
SynthSet — Synthetic Training Data with Isaac Sim Replicator
A domain-randomized synthetic-data factory built on Isaac Sim Replicator: procedurally varied lighting, textures and camera poses generate labeled datasets for a YOLO detector, deployed as a real-time ROS 2 perception node. The headline experiment quantifies the synthetic-to-real gap against a hand-labeled validation set — a continuation of the drone-imagery YOLO work for the City of Hamburg above.
Isaac Sim Replicator
YOLO
PyTorch
ROS 2
Python
Editor’s note:Synthetic data is one of the fastest-growing line items in robotics budgets; pairing it with real YOLO experience from a city government makes a rare, credible story.