Simulation Engineer, 1X

• Strong programming in Python, C++, or similar languages, with experience building environments or benchmarks using robotics simulators (MuJoCo, PyBullet, Isaac Sim, or equivalent)
• Demonstrated experience improving simulation fidelity and closing the sim-to-real gap for behavior that had to work on physical hardware
• Experience modeling sensors and their failure modes in simulation, including noise, bias, latency, and dropout
• Experience improving the performance of physics simulators or rendering pipelines
• Strong testing practices for simulation stacks used in robot learning, including tests that verify physical correctness and catch regressions
• 5+ years of relevant industry experience, or an advanced degree plus equivalent hands-on experience

Desirable

• The Simulation team builds the virtual environments and infrastructure that let 1X's AI team iterate on robot learning without being bottlenecked by real hardware. We construct physically realistic simulation worlds for NEO, scale synthetic data production, close the sim-to-real gap, and prototype new hardware virtually before it's manufactured
• Our work is a force multiplier for every research and development team at 1X: the faster and more faithfully we can simulate NEO's world, the faster the whole company learns
• Close the gap between how NEO behaves in simulation and how it behaves in the world. You will make the simulator physically accurate enough that results earned in simulation hold up on hardware, building and validating the dynamics, material, and sensor models that decide whether a simulated result means anything
• Every team that trains or evaluates against simulation is betting on the fidelity you deliver
• Key Outcomes:
• Physical Accuracy: Improve contact dynamics, articulated body models, and actuator and transmission fidelity, driven by system identification data from real hardware and kept current as hardware changes
• Compliant and Deformable Bodies: Model soft contact and structural compliance where rigid-body assumptions break down, so grasping and contact-rich behavior transfers instead of only working in simulation
• Sensing Realism: Model the sensing path as it actually behaves - noise, bias, dropout, latency, and sample timing - so a policy sees in simulation what it will see on the robot
• Measuring the Gap: Establish the benchmarks and real-robot comparisons that quantify sim-to-real error, so fidelity work is driven by evidence rather than intuition
• Randomize or Model: Decide which discrepancies to model faithfully and which to randomize over, with ranges grounded in the spread measured across the real fleet
• Determinism and Reproducibility: Make runs reproducible given the same inputs, so a regression is a real regression and any failure can be replayed and debugged
• Throughput and Environments: Optimize physics and rendering, and deliver the diverse environments and synthetic data pipelines that feed policy training and evaluation
• Physics Simulation Depth: Knows what makes a simulator physically accurate and computationally tractable, and how to tune contact dynamics and articulated body models for robot learning
• Sim-to-Real Judgment: Knows which differences between simulation and reality actually matter for policy transfer, and which can be absorbed by randomization
• Grounded in Real Data: Validates fidelity against measurements from real robots rather than trusting a simulation that looks right
• Fleet Thinking: Treats unit-to-unit variation across real robots as something to measure and model, not average away, so simulation reflects the fleet rather than one nominal unit
• Performance Engineering: Optimizes physics and rendering pipelines for throughput, and reasons explicitly about the tradeoff between fidelity and speed
• Rigorous Infrastructure: Writes tested, maintainable simulation code other teams depend on, treating determinism and correctness of the stack as first-class concerns
• Technical Direction: Sets the approach for a problem area, defends tradeoffs across fidelity, throughput, and effort, and partners with the teams that depend on simulation