Data-driven methods for design of model predictive controllers

Model predictive control (MPC) is a powerful feedback technique that is often used in data-driven robotics. The performance of data-driven MPC depends on the accuracy of the model, which often requires careful tuning. Furthermore, specifying the task with an objective function and synthesizing a feedback policy are not straightforward and typically lead to suboptimal solutions driven by trial and error. In this work, we seek to address these challenges by investigating data-driven methods for system identification, task specification, and control synthesis of unknown dynamical systems. First, we conduct a case study on the design of a data-driven MPC for performing automatic needle insertion in deep anterior lamellar keratoplasty, a challenging ophthalmic microsurgery task. We propose a data-driven method for controller synthesis and selection and demonstrate that the synthesized controller outperforms a state-of-the-art baseline in ex vivo physical experiments. Next, we present AutoMPC, an open-source Python package for automatic synthesis of data-driven MPC. We demonstrate the AutoMPC outperforms a state-of-the-art offline reinforcement learning algorithm on several standard control benchmarks. We further demonstrate that AutoMPC outperforms standard control baselines in physical experiments on an underwater soft robot.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2022

Author dc:creator

• Edwards, William

Contributors dc:contributor

• Hauser, Kris

• Robotics
• Model Predictive Control
• Automatic Tuning

Statement dc:rights

• Copyright 2022 William Edwards

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/117840

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F117840