Vision Engineer – Synthetic 3D Data (human)

Your Mission & Challenges

You will become part of our AI/Perception team and help shape the synthetic worlds in which our robots learn before entering the real world. You will be responsible for building scalable simulation and data generation pipelines, as well as developing World Action Models that enable humanoid and collaborative robots to understand, predict, and act within complex scenes.

Scene Simulation: Conception and construction of generated 3D scenes from real-world recordings for training and validating perception and policy models — from asset composition and physically accurate materials to dynamic population with humans and objects.

Domain Randomization: Development of systematic randomization strategies (lighting, textures, physics, sensor noise, camera poses) to close the sim-to-real gap and ensure robust generalization to real-world robot deployments.

3D Inpainting & Scene Editing: Development of methods for consistently adding, removing, and modifying objects in 3D scenes and point clouds — for data augmentation, occlusion handling, and runtime scene manipulation.

Simulation & Validation: Use of modern simulation environments (e.g. Isaac Sim, Gazebo, Unreal/Unity) and validation of real humanoid and collaborative robots — always with a focus on safety and robustness.

World Action Models: Design and fine-tuning of generative world models that predict scene dynamics, object interactions, and the consequences of robot actions — as a foundation for model-based planning and policy learning.

Interdisciplinary Collaboration: Work closely with AI, perception, control, and manipulation teams: define requirements, integrate sensor models, develop diagnostic tools, create documentation, and support CI/CD processes.

What We Can Look Forward To

Excellent Master's or PhD degree in Computer Science, Physics, Robotics, Computer Graphics, or a related field

3+ years of relevant experience in at least one of the core areas: scene simulation/compositing, synthetic data generation, 3D vision, or 3D augmentation

Strong programming skills in Python or modern C++; experience with CUDA is a plus

Hands-on experience with deep learning frameworks (PyTorch, TensorFlow) and generative models (diffusion models, NeRFs, Gaussian splatting, transformer-based world models)

Solid background in computer graphics, 3D geometry, and simulation: rendering pipelines, physical simulation, procedural generation, as well as experience with simulation environments such as Isaac Sim, Gazebo, Unreal, or Unity

Experience with synthetic data generation is advantageous

Experience with sim-to-real transfer and deployment on robotics hardware (e.g. NVIDIA Jetson) is a plus

Excellent problem-solving skills and the ability to work both independently and as part of a team

Very good written and spoken English