Javascript must be enabled for the correct page display

Torque-Controlled Grasping for the Unitree G1: Comparing Force-Impedance and Closed-Loop Control in Isaac Sim

Vinke, Jelle (2026) Torque-Controlled Grasping for the Unitree G1: Comparing Force-Impedance and Closed-Loop Control in Isaac Sim. Research Project, Industrial Engineering and Management.

[img]
Preview
Text
ResearchProjectJelleVinke.pdf

Download (5MB) | Preview

Abstract

This paper develops and validates an Isaac Sim–based manipulation platform for the Unitree G1 humanoid, aimed at enabling repeatable, torque-aware grasping experiments and generating trajectories for future learning- based approaches. The work first derives and verifies forward kinematics, geometric Jacobians, and a two-hand damped least-squares inverse kinematics solver for the G1 upper body. A gravity-compensated joint-space PD controller is then implemented and tuned to achieve stable end-effector motion. To close the perception–action loop, a simulated RealSense depth camera with semantic segmentation is used to estimate object pose and provide vision-driven targets for reaching and tracking. The core contribution is a comparative study of two finger grasp controllers for achieving a prescribed grasping torque. The first approach uses a force-impedance–inspired mapping that fits a static torque–position relationship per finger and inverts it to command joint positions. The second approach directly regulates measured joint torque using closed-loop feedback. Experimental results in simulation show that the force-impedance mapping exhibits substantial steady-state torque error under changing contact conditions, while the torque-feedback controller achieves accurate torque tracking with improved transient response and greater robustness in lifting tasks. Overall, the platform demonstrates closed-loop reaching, box tracking, and full pick–lift–place execution, providing a practical foundation for scalable data generation and future imitation or reinforcement learning for humanoid manipulation.

Item Type: Thesis (Research Project)
Supervisor name: Munoz Arias, M. and Carloni, R.
Degree programme: Industrial Engineering and Management
Thesis type: Research Project
Language: English
Date Deposited: 11 Mar 2026 08:26
Last Modified: 11 Mar 2026 08:26
URI: https://fse.studenttheses.ub.rug.nl/id/eprint/37252

Actions (login required)

View Item View Item