Embodied AI Real2Sim Engineer

Maxinsights•Santa Clara, CA

About The Position

Build the digital pipeline that transfers real-world robots, objects, and scenes into simulation. Reconstruct and calibrate their visual and physical properties, and deliver SimReady interactive assets to reduce the sim-to-real gap and support algorithm training, simulation evaluation, and policy transfer to real hardware. Own the end-to-end Real2Sim workflow using platforms such as Isaac Sim, MuJoCo, Genesis, and Blender. Work across 3D and appearance reconstruction, dynamic identification, and calibration of contact, friction, and inertia parameters to align simulated behavior and perception with real robots. Create, validate, and iterate SimReady assets using USD, URDF, or MJCF standards, configuring joints, collision shapes, mass and inertia, joint limits, semantic labels, and affordance properties. Process multimodal real-world capture data, including camera calibration, point clouds, meshes, textures, and PBR materials. Use methods such as 3D Gaussian Splatting (3DGS) and NeRF to digitize real scenes and export assets compatible with simulation engines. Develop automation for calibration, data processing, and asset quality inspection, and document asset specifications, workflows, and evaluation metrics. Collaborate with perception, robot learning, simulation engine, and hardware teams to design quantitative sim-to-real benchmarks, identify simulation bias, and improve calibration protocols and asset standards. Design system-identification excitation trajectories, collect time-series data from physical robots, and calibrate non-ideal dynamics such as nonlinear joint friction and transmission backlash.

Requirements

Proficiency in at least one simulation or graphics tool such as Isaac Sim, MuJoCo, Genesis, or Blender.
Solid understanding of computer graphics and rigid-body dynamics.
Familiarity with physics simulation engines such as PhysX and Newton.
Hands-on experience with Real2Sim pipelines and SimReady asset standards, including the ability to independently build, configure, and validate robot or object assets and tune their physical properties.
Knowledge of 3D reconstruction, camera imaging, point-cloud processing, and camera calibration.
Fundamental understanding of system identification, parameter calibration, and contact and friction models for physical simulation.
Familiarity with PBR rendering and material creation.
Ability to decompose requirements, implement technical solutions, run comparative experiments, and identify root causes of sim-to-real mismatches in visual reconstruction or physical simulation.
Ability to design dynamic system-identification experiments, analyze measured hardware data, and quantify sim-to-real errors caused by nonlinear dynamics.

Nice To Haves

Proven track record delivering Real2Sim projects, especially for robotic arms, dexterous hands, or humanoid robots.
Deep familiarity with USD, URDF, or MJCF and experience building asset libraries and automated asset-validation pipelines.
Proficiency in Python or C++ for calibration scripts, engine extensions, and data-processing tools.
Experience with 3DGS, NeRF, PBR reconstruction, and photogrammetry.
Background in system identification, excitation-trajectory design, dynamic-parameter estimation, and contact-model tuning.
Understanding of Sim2Real, reinforcement learning, and embodied AI closed-loop systems.
Experience with motion capture, multi-sensor calibration, and real-robot hardware experiments.
Familiarity with nonlinear friction models such as LuGre and Stribeck, and experience modeling and tuning joint hysteresis and backlash.

Responsibilities

Own the end-to-end Real2Sim workflow across simulation platforms, including 3D and appearance reconstruction, dynamic identification, and calibration of contact, friction, and inertia parameters.
Create, validate, and iterate SimReady assets according to USD, URDF, or MJCF standards.
Configure robot and interactive-object assets, including joints, collision shapes, mass and inertia, joint limits, semantic labels, and affordance properties.
Process multimodal real-world capture data, including camera calibration, point-cloud and mesh processing, texture processing, and PBR material calibration.
Use 3DGS, NeRF, and related reconstruction methods to digitize real scenes and export assets compatible with simulation engines.
Develop automated tools for calibration, data processing, and asset quality inspection.
Document asset specifications, calibration workflows, and evaluation metrics to improve pipeline throughput and sim-to-real alignment.
Collaborate with perception, robot learning, simulation engine, and hardware teams to design quantitative sim-to-real benchmark experiments, identify simulation bias, and improve calibration protocols and asset standards.
Design excitation trajectories for robot system identification, collect time-series dynamic data from physical robots, and calibrate and reproduce non-ideal dynamics, including nonlinear joint friction and transmission backlash.
Analyze measured hardware data and quantify sim-to-real errors introduced by nonlinear dynamics.