Robotics Engineer - Detection & Tracking; gn @ Space Defence Venture, Munich

About the venture

We are a space defence company. Our mission is to keep order in orbit.

Just as fighter jets secure the skies and navy ships secure the seas, we are building the systems that will protect the assets above us: the satellites our economies, security, and daily lives depend on. Space has become contested. As orbit fills with infrastructure and threats alike, Europe needs sovereign means to deter, defend, and uphold law and order beyond the atmosphere.

We are at the rare stage where the founding team becomes the company. Today, there are three founders building the first systems, setting the technical direction, and proving the mission. The next people to join will be among the company’s first employees, not joining an existing machine, but helping build one.

That means real authorship. The architecture, the engineering culture, the operating rhythm, and the standard of execution are still being formed. For the right person, this is the opportunity to help shape a European space defence company from its earliest days. We are looking for mission-driven people who want to defend the high ground, and who are ready to have honest debates about the use of force in space, its limits, and its ethics.

You will help build the autonomy architecture for a real-world robotic defence system.

That means perception, tracking, state estimation, and control working together in real time on edge hardware. Your work will determine whether the system detects a target, understands what it is seeing, estimates its own motion correctly, and keeps the effector locked on.

This is one of the first engineering roles. You will work directly with the three founders and have significant influence over core technical decisions from day one. You will not be handed a mature stack to maintain; you will help create the stack, test it in the real world, and push it toward operational use.

The work will be broad and practical. You might be tuning a detector, debugging a camera pipeline, chasing latency through ROS2 nodes, fixing bad timestamps, compensating for ego-motion, calibrating sensors, or running field tests with hardware that fails in new ways every time.

The expectation is not that you know everything. It is that you have seen enough real systems to know where they break, and that when they do, you can figure it out fast.

You will own the robotics stack from perception through to actuation, including:

• Real-time object detection, classification, and multi-object tracking
• State estimation and sensor fusion across GPS, IMU, compass / magnetometer, and motor encoders
• Camera geometry, calibration, coordinate frames, and ego-motion compensation
• ROS2 integration connecting perception, estimation, and pan-tilt control
• Edge deployment, debugging, and iteration on Jetson-class hardware

You will build for real constraints: latency budgets, live camera streams, imperfect sensors, outdoor field conditions, and fast field-test cycles.

You will also help create the foundations around the stack: deployment workflows, data pipelines, evaluation tools, live telemetry, and the first steps toward space-payload operational requirements.

We are looking for someone with strong practical depth in robotics, computer vision, or autonomy.

You should bring:

• Production-level OpenCV, plus strong Python and C++
• ROS2 experience in real systems, not only tutorials or coursework
• Embedded or real-time camera pipeline experience, ideally with GStreamer or similar
• Practical experience with object detection, tracking, and live perception systems
• Hands-on state estimation experience, including EKF, UKF, or equivalent approaches
• A good feel for latency, threading, buffering, dropped frames, timestamps, and real-time system failure modes.

You must have NATO member state citizenship.

A camera feed that worked on the bench but failed in the field.

A detector that looked good offline but missed targets live.

A tracker that broke under fast motion, occlusion, vibration, low contrast, or ego-motion.

An IMU, compass, or GPS signal that looked reasonable until it was mounted on real hardware.

A field test where the system failed live and you had to decide what to measure, patch, or rebuild first.

Experience with TensorRT, ONNX Runtime, Jetson Orin, CUDA, Docker on embedded hardware, robot localization, GNSS / RTK, NED / ENU frames, visual-inertial odometry, behaviour trees, state machines, or IR / thermal / near-IR imaging.

Experience with small, fast-moving, or low-contrast targets such as drones, birds, aircraft, or distant moving objects is especially relevant.