Overview

A Sim2Real case verifying whether low-level control gains tuned in Isaac Sim remain valid on a real robot.

Project Background & Purpose

This video verifies whether ADRC (Active Disturbance Rejection Control) gains tuned in NVIDIA Isaac Sim hold up when applied directly to a real robot. ADRC is a low-level control method that computes joint torques; WIM implemented it and ran the identical algorithm in both Isaac Sim and on the real robot. Control gains tuned in simulation often behave differently on a real robot due to unmodeled factors like friction and backlash, requiring significant on-site retuning time.

Components

• ADRC (Active Disturbance Rejection Control) algorithm
• Extended State Observer (ESO)
• W-RC AI robot controller
• NVIDIA Isaac Sim

Process Steps

1. 1Tune ADRC gains in Isaac Sim
2. 2Apply the identical ADRC algorithm to the real robot via the W-RC controller
3. 3The ESO corrects sim-to-real mismatch in real time while motion is verified