Neura Robotics

Cognitive robots with perception, manipulation, collaboration

Dynamics – System Identification & Estimation Engineer - Pla, Mensch

Robotics & Autonomous SystemsConfirmed live in the last 24 hours

Full-timeNo salary listedIn-person

About the Job

Requirements

A Master of Science or PhD in robotics, mechanical engineering, electrical engineering, or a similar field, with a strong focus on dynamics, estimation, and control.
At least 4 years of experience with state estimation or system identification for real-time robots using physical hardware, not only simulation.
Experience performing system identification on physical robotic hardware, including trajectory design, least-squares or maximum-likelihood fitting, and actuator and transmission identification.
Experience implementing state estimation with an extended Kalman filter or unscented Kalman filter for floating-base pose, velocity, and contact states on legged or mobile platforms.
Deep knowledge of rigid-body dynamics, including contact modeling, actuator behavior, and how model errors can cause instability.
Experience supporting control systems such as model predictive control or whole-body control, or learned policies such as reinforcement learning, on physical hardware, and understanding how model quality affects transfer.
C++ skills for production real-time systems and Python skills for analysis, tooling, and calibration pipelines.

Responsibilities

Define model-accuracy metrics and own the gap between simulation and real hardware behavior, particularly during dynamic movements and contact-rich interactions.
Identify system parameters on 4NE1 hardware, including motor constants, joint friction, and transmission dynamics; design excitation trajectories, fit regressors, analyze observability, and iteratively optimize using real hardware data.
Author and maintain MuJoCo and Isaac Sim models that reproduce real behavior under load and contact; parameterize contact models and calibrate actuator models.
Implement and tune extended Kalman filters or unscented Kalman filters for pose, velocity, and contact states (floating base) in a real-time loop, providing input for controllers and locomotion or manipulation policies.
Develop sim-to-real parameter-estimation loops and data-driven calibration from hardware; validate using motion capture or external references and maintain a continuous feedback loop between hardware tests and simulation.
Debug failures caused by inaccurate models, including instability from incorrect dynamics, drift or bias in state estimation, and faulty contact or force models.
Provide updated Pinocchio model parameters to whole-body-control and state-estimation engineers; coordinate trajectories with locomotion and reinforcement-learning or control teams.

Desired Qualifications

Hands-on experience with humanoid or legged robots, especially bipedal dynamics.
Experience with differentiable simulators for gradient-based system identification, such as Brax or DiffTaichi.
Experience with sim-to-real transfer, including domain randomization, adaptive calibration, and residual physics models.
Experience with Pinocchio for rigid-body models and sensitivity analyses.
Experience authoring MuJoCo models, including MJCF contact parameters, actuator models, and tendon dynamics.
Experience with factor-graph-based estimation, such as GTSAM or iSAM2, for IMU and kinematics fusion.
Publications or open-source contributions in legged robot dynamics, system identification, or sim-to-real.

Deine Mission & Challenges

Dynamic Model Accuracy Ownership: Du definierst Metriken für Modellgenauigkeit und verantwortest die Lücke zwischen Simulation und realem Hardware-Verhalten – besonders bei dynamischen Bewegungen und kontaktreichen Interaktionen
System Identification auf 4NE1 Hardware: Motor-Konstanten, Gelenk-Reibung, Getriebedynamik – Design von Anregungstrajektorien, Regressor-Fitting, Observability-Analysen, iterative Optimierung anhand echter Hardware-Daten
Simulation Model Authoring & Maintenance: MuJoCo- und Isaac-Sim-Modelle, die reales Verhalten unter Last und Kontakt abbilden; Parametrisierung von Kontaktmodellen, Kalibrierung von Aktuator-Modellen
Real-Time State Estimation: Implementierung und Tuning von EKF/UKF für Pose, Velocity und Kontaktzustände (Floating Base) im Echtzeit-Loop; Input für Controller und Loco-/Manipulations-Policies
Sim-to-Real Pipeline: Parameter-Estimation-Loops, datengetriebene Kalibrierung aus Hardware, Validierung via Motion Capture oder externe Referenzen – kontinuierlicher Feedback-Loop zwischen Hardware-Tests und Simulation
Failure Mode Ownership: Debugging von Fehlern aufgrund ungenauer Modelle – z. B. Instabilität durch falsche Dynamik, Drift/Bias in der Estimation oder fehlerhafte Kontakt-/Kraftmodelle
Cross-Team Interface: Bereitstellen von aktualisierten Pinocchio-Modellparametern für WBC- und State-Estimation-Engineers; Abstimmung von Trajektorien mit Locomotion- und RL/Control-Teams

Auf was können wir uns freuen

MSc oder PhD in Robotik, Maschinenbau, Elektrotechnik oder ähnlichem – mit starkem Fokus auf Dynamik, Estimation und Regelung

4+ Jahre Erfahrung mit State Estimation oder System Identification für Echtzeit-Roboter – auf echter Hardware, nicht nur Simulation
System Identification auf physischer Robotik-Hardware: Trajektorien-Design, Least-Squares oder Maximum-Likelihood Fitting, Identifikation von Aktuatoren und Transmissionen
State Estimation Implementierung: EKF oder UKF für Floating-Base Pose, Velocity und Kontaktzustände auf legged oder mobilen Plattformen
Tiefe Kenntnisse in Rigid Body Dynamics: Kontaktmodellierung, Aktuatorverhalten und wie Modellfehler zu Instabilitäten führen
Erfahrung in der Unterstützung von Control-Systemen (MPC, WBC) oder gelernten Policies (RL) auf echter Hardware – Verständnis, wie Modellqualität den Transfer beeinflusst
C++ für produktive Echtzeit-Systeme; Python für Analyse, Tooling und Kalibrierungs-Pipelines

Nice to have

Hands-on Erfahrung mit Humanoiden oder legged Robots – 4NE1 ist ein Full-Size Humanoid, bipedale Dynamik ist highly relevant
Differenzierbare Simulatoren für gradientenbasierte System Identification (z. B. Brax, DiffTaichi oder ähnlich)
Erfahrung mit Sim-to-Real Transfer: Domain Randomization, adaptive Kalibrierung, Residual Physics Models
Pinocchio für Rigid-Body-Modelle und Sensitivitätsanalysen
MuJoCo Model Authoring: MJCF Kontaktparameter, Aktuator-Modelle, Tendon-Dynamiken
Factor-Graph-basierte Estimation (GTSAM, iSAM2) für IMU + Kinematics Fusion
Publikationen oder Open-Source-Beiträge in legged Robot Dynamics, System Identification oder Sim-to-Real

About the Company

Neura Robotics

Cognitive robots with perception, manipulation, collaboration

Neura Robotics builds cognitive robots for industrial and service settings, combining perception, manipulation, and human collaboration. It develops both the hardware and the software stack for Physical AI, creating robot platforms like MAiRA, MAV, LARA and MiPA along with training infrastructure. The system works by integrating sensing, decision-making, and manipulation capabilities on a hardware-software stack that can be deployed in large-scale production, with the company planning robots fleets and NEURA Gyms for ongoing training and improvement. What sets Neura Robotics apart is its end-to-end approach to Physical AI, including a focus on mass production, an established global deployment plan, and a broad ecosystem of strategic backers and partners that support scaling and international expansion. The goal is to enable worldwide deployment of its robot platforms and associated training infrastructure, turning physical AI into widely available industrial and service automation.

Company size: 501-1,000
Company stage: Series C
Total funding: $1.7B
Headquarters: Metzingen, Germany
Founded: 2019

Recently Posted Jobs

Explore more opportunities at Neura Robotics

See more jobs