Dynamics, System Identification & Estimation Engineer - PLA (human)

Dynamic model accuracy ownership: defining model fidelity metrics and owning the gap between simulation behaviour and real-hardware behaviour across dynamic motion and contact-rich interactions

System identification on 4NE1 hardware: motor constants, joint friction, transmission dynamics — excitation trajectory design, regressor fitting, observability analysis, iterative refinement against hardware data

Simulation model authoring and maintenance: MuJoCo and Isaac Sim models that match real-world behaviour under dynamic loading and contact; contact model parameterisation, actuator model calibration

Real-time state estimation: floating-base EKF/UKF implementation and tuning for pelvis pose, velocity, and foot contact state at RT loop rates; feeds downstream controllers and loco-manipulation policy inputs

Sim-to-real pipeline: parameter estimation loops, hardware-data-driven calibration, validation against motion capture or external reference systems — the continuous feedback loop between hardware campaigns and updated sim models

Failure mode ownership: debugging model-accuracy-driven failures — control instability from inaccurate dynamics, estimation drift or bias causing divergence, incorrect contact/force estimation leading to instability in dynamic interactions

Cross-team interface: supplying updated Pinocchio model parameters to the WBC and State Estimation Engineers in Core Robot Software; aligning on excitation trajectory designs with the Locomotion and RL/Control Engineers

MSc or PhD in Robotics, Mechanical Engineering, Electrical Engineering, or a related field with a strong foundation in dynamics, estimation, and control

4+ years of experience developing state estimation or system identification solutions for real-time robotic systems — on real hardware, not simulation-only

System identification on physical robotic systems: excitation trajectory design, least-squares or maximum-likelihood regressor fitting, actuator and transmission parameter identification

State estimation implementation: EKF or UKF for floating-base pose, velocity, and contact state on a legged or mobile robot platform

Rigid body dynamics depth: contact modelling, actuator behaviour, and how model inaccuracies propagate to control instability — not just theoretical familiarity

Experience supporting control systems (MPC, WBC) or learned policies (RL) through hardware deployment — understanding how model quality gates policy transfer

C++ for production RT systems; Python for analysis, tooling, and calibration pipelines

Humanoid or legged robot hands-on experience — 4NE1 is a full-size humanoid; bipedal dynamics and contact complexity are directly relevant

Differentiable simulators for gradient-based system identification (Brax, DiffTaichi, or comparable)

Sim-to-real transfer methodology: domain randomisation, adaptive calibration, residual physics modelling

Pinocchio for rigid-body model computation and parameter sensitivity analysis

MuJoCo model authoring: MJCF contact parameters, actuator models, tendon dynamics

Factor graph-based estimation (GTSAM, iSAM2) for tightly-coupled IMU + kinematics fusion

Publications or open-source contributions in legged robot dynamics, system identification, or sim-to-real transfer