REAL

Practical method to complete Learning Model Predictive Control with generalization capability

Török, Ferenc and Péni, Tamás (2020) Practical method to complete Learning Model Predictive Control with generalization capability. In: IFAC World Congress, 2020.07.11-2020.07.17, Virtual (Berlin, Germany).

[img] Text
Török-Practical method to complete Learning Model Predictive Control with generalization capability-IFAC20.pdf - Accepted Version
Restricted to Repository staff only

Download (379kB) | Request a copy
Official URL: https://www.ifac2020.org

Abstract

The paper presents a practical method to complete Learning Model Predictive Control (LMPC) with generalization capability. LMPC has been developed by F. Borrelli and his co-authors for systems performing iterative tasks. The method is based on saving the state trajectories of successful runs and using this database to improve the control performance in the future iterations. When the controller faces a new task, the database is cleared and the learning phase starts over. This paper addresses the question of how a general knowledge base can be built to warm start the learning process. As a potential solution, a practical method is proposed. The algorithm is tailored specifically to the autonomous racing application, but the concept can be extended to a wider class of control problems. The procedure includes the construction of special teaching tracks, on which the trajectory database is generated and a multi-step migration procedure for transferring the learned trajectories onto any new track. The efficiency of the method is demonstrated by numerical simulations.

Item Type: Conference or Workshop Item (Paper)
Subjects: T Technology / alkalmazott, műszaki tudományok > T2 Technology (General) / műszaki tudományok általában
Depositing User: Péni Tamás
Date Deposited: 24 Sep 2020 10:31
Last Modified: 24 Sep 2020 10:31
URI: http://real.mtak.hu/id/eprint/113879

Actions (login required)

Edit Item Edit Item