Rapid Deployment of Model Predictive Control for Robotic Systems: From IMPACT to ROS 2 Through Code Generation

This paper presents an approach for automatic code generation of nonlinear model predictive controllers (NMPC) for robotic systems as ROS 2 packages. The developed approach extends IMPACT - a framework for NMPC specification, prototyping and deployment - to expand its usability towards the robotics community. With such extension, the wide variety of numerical optimization solvers interfaced within IMPACT are made available to any project involving a ROS 2 architecture. The extensions to IMPACT are based on three submodules that automatically generate robot models from robot description files, generate ROS 2 packages for NMPC-based robot control, and benchmark the computational complexity of generated functions and algorithms. A hardware-in-the-loop simulation is presented for tunnel-following NMPC on a 7-degrees-of-freedom robot manipulator, illustrating the workflow from problem definition to deployment on embedded hardware.