Whole-Arm Manipulation

Multi-contact feedback model predictive control (MPC) leveraging whole-arm sensing, and extending towards manipulation policy learning.

Whole-Arm Manipulation

This project focuses on Whole-Arm Manipulation, progressing from sensing-in-the-loop control to learning-based manipulation policies. By integrating real-time contact estimation with feedback control loops (MPC) and scaling to human-to-robot retargeting, our research aims to enable robot arms to perform complex, contact-rich physical interactions and collaborative manipulation tasks safely and robustly.

Key Research Topics

  • Multi-Contact Feedback MPC: Incorporating real-time contact location and force feedback directly into the Model Predictive Control (MPC) loop to dynamically react to physical disturbances and environment changes.
  • Whole-Arm Retargeting: Developing human-to-robot motion retargeting pipelines to transfer human whole-arm movements onto robot manipulators, establishing a high-fidelity data collection tool to facilitate policy learning.
  • Whole-Arm Manipulation Policy: Training learning-based manipulation policies (e.g., via imitation learning or reinforcement learning) that coordinate the kinematics and contact forces of the entire robot arm for complex, contact-rich tasks.
  1. Online Multi-Contact Feedback Model Predictive Control for Interactive Robotic Tasks (ICRA 2024)
  2. Optimization-based Multi-Contact Force Control (KRoC 2024)
    • Seo Wook Han, Maged Iskandar, Jinoh Lee, Min Jun Kim