Skip to main content

Building a Physics-Accurate Digital Twin of G1

Duration: 60 min · Level: Intermediate · Module: 7. Simulation & Digital Twins · Focus: digital-twin, URDF, Isaac-Sim, modeling

Learning objectives

By the end of this lesson you will be able to explain and apply:

  • URDF → USD pipeline
  • Inertial parameters
  • Actuator modeling
  • Sensor simulation
  • State estimation bridge

Why this matters​

A digital twin is a live, synchronized simulation of the real robot — updated in real-time from sensor data.

Overview​

A digital twin is a live, synchronized simulation of the real robot — updated in real-time from sensor data. For G1 development, the digital twin serves three purposes: visualization, predictive maintenance, and policy pre-testing before hardware deployment.

Key concepts​

Key idea

URDF → USD pipeline: convert G1's URDF (Unified Robot Description Format) to USD for Isaac Sim; NVIDIA provides isaac_ros_urdf for automated conversion

  • Inertial parameters: measured vs nominal inertia tensors for each link differ by 5-20%; use system identification (swing experiments) to measure actual parameters
  • Actuator modeling: add motor dynamics to simulation (current limiting, back-EMF, thermal derating model) for accurate torque prediction
  • Sensor simulation: add realistic noise models for each sensor based on manufacturer datasheets; IMU noise, camera calibration errors, encoder quantization
  • State estimation bridge: real-time pose estimate from onboard SLAM → update digital twin state; latency <50ms for useful synchronization
  • Use cases: predict joint wear patterns, test new locomotion policies in digital twin before hardware deployment, train maintenance technicians on correct repair procedures

Check your understanding​

Try to recall each answer before expanding it.

Q1. What do you know about URDF → USD pipeline?

convert G1's URDF (Unified Robot Description Format) to USD for Isaac Sim; NVIDIA provides isaac_ros_urdf for automated conversion

Q2. What do you know about Inertial parameters?

measured vs nominal inertia tensors for each link differ by 5-20%; use system identification (swing experiments) to measure actual parameters

Q3. What do you know about Actuator modeling?

add motor dynamics to simulation (current limiting, back-EMF, thermal derating model) for accurate torque prediction

Q4. What do you know about Sensor simulation?

add realistic noise models for each sensor based on manufacturer datasheets; IMU noise, camera calibration errors, encoder quantization

Q5. What do you know about State estimation bridge?

real-time pose estimate from onboard SLAM → update digital twin state; latency <50ms for useful synchronization

← Previous: 7.2 Domain Randomization: The Bridge from Sim to Real

Part of Module 7: Simulation & Digital Twins.