AI and Computer Vision Engineer

Meyandy LLC; München; Hybrid; EUR 90.000 - 130.000; Vollzeit

Zusammenfassung

The Exploration Company in Munich is seeking an experienced AI and Computer Vision Engineer to build the perception capability behind autonomous close-proximity operations: estimating relative pose from camera data, largely using synthetic and lab data, and optimized for flight hardware.

You will design, train and evaluate deep-learning models, manage data pipelines, and adapt models for embedded and neuromorphic hardware while prioritizing reliability and explainability.

Qualifikationen

• Degree (MSc or PhD) in CS/EE/ robotics with ML focus.
• 3+ years building DL CV systems incl. pose estimation and 3D perception.
• Experience with sim-to-real and model optimization techniques.

Aufgaben

• Design, train and evaluate deep-learning models for 6-DoF pose estimation.
• Own training pipeline: dataset generation, augmentation, domain adaptation, experiment tracking.
• Optimize models for embedded/neuromorphic hardware; balance latency and power.
• Port models to hardware and validate accuracy with energy constraints.
• Run validation campaigns on hardware-in-the-loop testbeds.

Kenntnisse

Python

PyTorch

3D geometry

Linux

Git

Experimentation

Ausbildung

MSc/PhD in CS/EE/Robotics

Tools

CUDA

GPU clusters

Containerization

Jobbeschreibung

Here at The Exploration Company, we are building innovative aerospace technologies that advance the future of space transportation.We want you as a hands-on AI and Computer Vision Engineer to build the perception capability behind autonomous close-proximity operations: models that estimate the relative position and orientation of non-cooperative spacecraft from camera images, trained largely on synthetic and lab data, and optimized to run on the compute we can actually fly.This is a builder role. You write the training code, run the experiments, take the models onto embedded and neuromorphic hardware, and own the results.## Key ResponsibilitiesIn your capacity as AI and Computer Vision Engineer, your role will be continuously evolving, but day to day your duties will include:* Designing, training and evaluating deep-learning models for 6-DoF pose estimation of non-cooperative spacecraft* Owning the full training pipeline: dataset generation and management, augmentation, domain adaptation between synthetic, laboratory and orbital imagery, experiment tracking and reproducibility.* Optimizing models for flight-representative compute (knowledge distillation, pruning, quantization and quantization-aware training) and benchmarking latency, memory and power against onboard constraints.* Porting and evaluating models on embedded and neuromorphic hardware, and characterizing the accuracy versus energy trade-off.* Building explainability and uncertainty into the pipeline so failure modes such as high occlusion can be debugged and the technology is a credible candidate for certification.* Exploring privacy-preserving and distributed training approaches that let us improve models with partners without exchanging raw data.* Prototyping lightweight self-supervised refinement methods for later in-flight model adaptation on unlabeled imagery.* Defining requirements, test scenarios and validation criteria together with GNC/FPO, and supporting the selection and characterization of space-qualified camera sensors.* Running validation campaigns on hardware-in-the-loop testbeds and analyzing the results.* Managing our training compute footprint across cloud GPU and internal HPC efficiently.## What we would love to see from youIn this role, ideally, you will have the following:**Education*** Degree (MSc or PhD) in computer science, electrical engineering, robotics, aerospace, physics, or a comparable field with a strong machine learning focus.**Experience*** 3+ years building and shipping deep-learning computer vision systems such as object detection, keypoint detection, pose estimation, or 3D perception. PhD work in the field counts.* Demonstrable experience taking a model from research prototype to a constrained target: quantization, distillation, latency optimization, deployment on embedded or accelerator hardware.* Experience training on synthetic data and dealing with the sim-to-real gap.* Hands-on lab work: cameras, calibration, test setups, collecting and annotating your own data.**Skills and Competencies*** Strong Python and PyTorch (or JAX/TensorFlow); clean, version-controlled, reproducible code.* Solid classical computer vision and 3D geometry: camera models, intrinsics and extrinsics, distortion, PnP, RANSAC, coordinate frames.* Comfortable with Linux, Git, containers, and running large training jobs on GPU clusters or in the cloud.* Genuinely hands-on and self-directed: you will carry your work largely on your own, with review support rather than daily direction.* Able to communicate results clearly in technical reports and reviews.* Working proficiency in English; German is a plus.**Nice to have*** Familiarity with spacecraft rendezvous, docking, or vision-based navigation, and with benchmarks such as SPEED/SPEED+ and the ESA pose estimation challenges.* Spiking neural networks and neuromorphic hardware, or event-based cameras.* Federated learning, differential privacy, or secure agg...