Research Engineer, Robotics: Physics Simulation and Sim-to-Real
The Allen Institute for Artificial Intelligence6 months ago
Base Salary
$129k - $193k/yr
Responsibilities
- Design, implement, and improve contact, friction, and constraint solvers.
- Diagnose and eliminate simulation instability, including jitter, interpenetration, tunneling, energy gain, constraint drift, and GPU non-determinism.
- Make articulated robots and closed kinematic loops stable across MuJoCo, Newton, and PhysX, and contribute fixes upstream to open-source engines.
- Select and evaluate integrators, timesteps, and substeps for accuracy and throughput in large-scale GPU-batched training.
- Build stress and regression tests for long-horizon, randomized, and contact-rich physics scenarios.
- Prepare simulation-ready assets with accurate collision geometry, inertial properties, joint limits, and contact settings across URDF, MJCF, and OpenUSD.
- Fit simulation parameters to real robot data, including friction, inertia, payload, actuator gains, and latency.
- Calibrate contact behavior against recorded pushes, slides, grasps, and handovers.
- Maintain a sim-to-real correlation suite that replays trajectories and measures joint, velocity, lag, and contact-force errors.
- Use measured uncertainty to define domain-randomization ranges and assess whether simulated policy success predicts real-world performance.
- Document robot and scene parameters, their sources, and their reliability in model cards.
Requirements
- PhD in robotics, mechanical engineering, applied mathematics, physics, computer graphics, or a related field; alternatively, a master’s degree plus three years of experience, or an exceptional bachelor’s candidate with three or more years of relevant industry experience.
- Demonstrated depth in rigid-body dynamics, constrained multibody systems, articulations, contact and friction modeling, and numerical integration of stiff systems.
- Experience writing solver or physics-engine code in a research codebase, production engine, or simulator such as MuJoCo, Newton/Warp, PhysX, Drake, Bullet, or Genesis.
- Track record diagnosing and fixing simulation instability with an understanding of the underlying causes.
- Experience fitting simulation parameters to measured data and quantitatively validating results against real-world recordings.
- Proficiency in Python and C++.
- Experience with robot description formats including URDF, MJCF, and OpenUSD.
- Knowledge of at least one major simulation stack at source level.
- Preferred: GPU physics programming with CUDA, NVIDIA Warp, or JAX and experience scaling simulation for reinforcement or imitation learning.
- Preferred: deformable, cable, or FEM simulation, differentiable simulation, Bayesian or gradient-based parameter estimation, learned residual dynamics, robot manipulation, grasping, publications, or open-source contributions.
Benefits
- Medical, dental, vision, life, disability, and employee assistance program coverage for team members and families.
- Voluntary life and accidental death and dismemberment insurance, health savings and reimbursement accounts, and healthcare and dependent care flexible spending accounts.
- 401(k) plan.
- $125 per month for commuting or internet expenses and $200 per month for fitness and wellbeing expenses.
- Up to ten sick days, seven personal days, 20 vacation days, and twelve paid holidays annually.
- Annual bonuses and participation in a long-term incentive plan.
- Expected to work from the Seattle office; on-site requirements vary by position and team.
- Ongoing education, guest speakers, learning opportunities, family leave, and a stated emphasis on work/life balance.
About The Allen Institute for Artificial Intelligence
The Allen Institute for Artificial Intelligence (AI2) is a Seattle-based nonprofit research institute that advances AI through open models, datasets, and tools, and runs applied projects in NLP, vision, robotics, and conservation. It creates widely used resources such as Semantic Scholar and AllenNLP, and operates the AI2 Incubator to spin out startups. Founded in 2014 by Paul Allen, it partners with academia and industry and publishes open research.