Learning Shape Control of Multi-Agent Systems with Lagrangian Neural Networks

GAMONAL FERNANDEZ, MANUELA; MORENO, PATRICIO; COLOMBO, LEONARDO J.

Spokane, Washington

Conferencia; SIAM Conference on Control and Its Applications (CT21); 2021

Society for Industrial and Applied Mathematics

Shape control of double integrator agents can be seen as a stabilization system whose evolution can be described by forced Euler-Lagrange equations. If agents are subject to unknown disturbances, desired shapes can not be achieved with the classical controllers. We propose a Neural Network for forced Lagrangian systems to learn the unknown disturbances, and we use the learning to re-design the controller to achieve the desired shape. A numerical example highlights the effectiveness of the proposed learning-based control law.