Location:
Anywhere, Worldwide | Madrid, Spain
Our client is looking for a Simulation Fidelity & Evaluation Engineer to own sim-to-real transfer validation — measuring whether changes in simulation fidelity actually move real-world model performance, and building the machinery that produces that answer.
Tech Stack
- Rigid-body physics engines
- Camera and LiDAR sensor modeling, domain randomization
- Real-time rendering engines / scene-description formats
- Python
- Reproducible-evidence workflows (pinned inputs, sealed protocols, frozen scenario sets)
Key Responsibilities
- Design and run end-to-end fidelity-to-transfer experiments — hypothesis and failure condition defined upfront, confounds addressed, results reported honestly
- Qualify measurement instruments before trusting them to produce results
- Own the evaluation runtime and behaviour metrics
- Extend camera and LiDAR sensor models
- Work the rendering path where needed
- Maintain a reproducible-evidence workflow so results can be rerun by others
Mandatory Requirements
- Has personally designed and run an experiment measuring whether a simulation fidelity change affected real-world model performance — including one that produced a negative result. Doesn't need to be robotics-specific. (This is the top screening criterion.)
- Strong experimental design and applied statistics — pre-defined failure criteria, confound identification, instrument validation, appropriate skepticism toward single metrics
- Perception model training and evaluation — detection metrics, evaluation harnesses, distinguishing signal from noise
- Comfortable working directly in the render and sensor path
- Clear technical writing that lets others reconstruct reasoning and reproduce results
Big Pluses
- Hands-on depth in a rigid-body physics engine; experience owning an evaluation/benchmarking harness others depended on
- LiDAR and/or camera sensor modeling; domain randomization experience
- Experience with a real-time rendering engine or scene-description format
- Python, with experience keeping evidence reproducible over time
Alternative Backgrounds
Strong fits have come from autonomous vehicles, synthetic data, model evaluation, academic sim-to-real research, or building benchmarking/evaluation platforms others relied on. Industry background matters less than having a genuine experimental track record.
Interested in learning more? Apply and let's talk!
