Simulation Engineer

About the Role

The WoRV team develops Mobile Manipulation VLA models where a single model performs both navigation and manipulation in brownfield environments not designed for robots — farmland, factories, construction sites.

This position designs and builds, to a product-grade level, the simulation stack that connects everything from VLA model training-data generation to performance evaluation. The goal is not a throwaway environment for a single experiment, but a Safe, Reliable, and Scalable simulation capability that can be reused across many tasks, robots, and field sites.

What You'll Do

• Design and develop simulation environments for VLA model training and evaluation (implementing manipulation / wheeled robots, configuring task scenarios and environmental conditions)
• Reduce the sim-to-real gap and improve model generalization through physics parameter approximation, sensor noise modeling, domain randomization, and automated object/environment generation
• Build simulation-based pipelines, benchmarks, and evaluation harnesses that connect training-data generation to quantitative evaluation
• Analyze discrepancies between real-robot and simulation results to improve simulation fidelity, reproducibility, and throughput
• Collaborate with the AI / Control / Data / PM teams to turn real customer-site problems into reproducible simulation scenarios

Requirements

• You can directly implement and debug the functionality you need in Python / C++
• You can read code to understand structure and behavior even when documentation is limited
• You can quickly learn and apply new simulators and frameworks
• You can weigh trade-offs not only in accuracy but also in reproducibility, throughput, and maintainability

Preferred Qualifications

• Experience with robotics simulators such as Isaac Sim or MuJoCo
• Project experience in robot manipulation or navigation
• Experience with domain randomization and sim-to-real transfer
• Experience building evaluation environments for Physical AI / VLA
• Experience with large-scale data generation, parallel simulation, and rendering/physics-engine optimization

Who We're Looking For

• Someone who understands the difference between "works in simulation" and "works on the real robot," and is motivated to close that gap
• Someone who wants to build a reusable simulation capability across many tasks and field sites, not a one-off environment
• Someone who learns new technologies quickly but can decisively simplify when a simpler approach is the right one