Real-Time Deformable-Contact-Aware Model Predictive Control for Force-Modulated Manipulation

This code is primarily uses DDP as the optimizer with autodiff for real-time implementation. The analytical model is derived using RobCoGen. control-toolbox is used to use Autodiff functionality for real-time implentation.
This repo uses DDP (Differential Dynamic Programming) + Soft-Contact inside a ADMM (Alternating Direction Multiplier Method) outer loop.

Structure

• Standalone DDP code (C++) with a soft contact model
• ADMM integration with constraints and DDP
• Model Predictive Control (MPC)

Dependencies

Dynamics of the robot are computed using

• orocos KDL
• control-toolbox

Installation

mkdir build & cd build
ccmake ..

To use the eigen library installed in the system, enable the the option USE_SYSTEM_EIGEN. To build from source.

make

How to use

• Scripts for executables can be found in src

• Files specific to ADMM - admm

• Files specific to DDP - ddp

• Following executables are generated

  • admm-contact (admm with orocos KDL dynamics, finite difference)
  • admm-contact-rcg (admm with RobCoGen dynamics, AutoDiff)
  • admm-mpc-contact - (MPC - admm with orocos KDL dynamics, finite difference)
  • admm-mpc-contact-rcg - (MPC - admm with RobCoGen dynamics, AutoDiff)

Aknowledgement and Contact

Lasitha Wijayarathne - lasitha@gatech.edu

Ziyi Zhou - zhouziyi@gatech.edu

Citation

@article{wijayarathne2023real,
  title={Real-Time Deformable-Contact-Aware Model Predictive Control for Force-Modulated Manipulation},
  author={Wijayarathne, Lasitha and Zhou, Ziyi and Zhao, Ye and Hammond, Frank L},
  journal={IEEE Transactions on Robotics},
  year={2023},
  publisher={IEEE}
}