Maik Pfefferkorn M.Sc.
Working area(s)
Machine-Learning-supported Model Predictive Control, Gaussian Processes (GPs) for Systems Modeling and Control, Uncertainty Propagation and Safety Guarantees in Gaussian-Process-based Model Predictive Control, Stochastic Model Predictive Control, Control of Scanning Quantum Dot Microscopy
Contact
maik.pfefferkorn@iat.tu-...
work +49 6151 16-25171
fax 06151 16-25172
Work
S3|10 520
Landgraf-Georg-Str. 4
64283
Darmstadt
Many control and automation problems are subject to safety-critical constraints and performance requirements that need to be satisfied at all times. Providing such guarantees is, in general, a challenging task. Model predictive control, a nowadays widely used, advanced, optimization-based control method, is capable of providing the required guarantees. However, to this end, it relies on the availability of high-quality system models and (almost) perfect knowledge of the system and the environment. In the presence of process uncertainties and disturbances, the closed-loop performance is significantly deteriorated and guarantees become conservative.
It is thus important to understand the effect of imperfect knowledge of the system or its environment on the closed-loop performance. We focus on dynamic systems that are subject to additive stochastic uncertainty and dedicated stochastic model predictive control formulations that account for this uncertainty. However, the probability distribution of the uncertainty is often not (exactly) known in practice. Instead, one often designs model predictive controllers that are robust against the particular shape of the distribution given higher-level knowledge about the distribution, e.g., knowledge of low-order moments. Here, the research objective is to understand and quantify the performance loss of distributionally robust model predictive controllers compared to their counterparts with exact knowledge of the distribution.
Besides understanding the performance loss due to uncertainty and disturbances, one can aim to compensate for it. To this end, we aim to examine and develop model predictive control approaches that provide safety guarantees and restore closed-loop performance despite uncertainty. In particular, we combine model predictive control and Gaussian process regression, a probabilistic machine learning method, to improve the control performance by learning from data. This raises several research questions, which we aim to tackle: How to combine Gaussian processes and model predictive control in a computationally efficient way? How to cope with uncertainty introduced through the machine learning algorithm? How to derive closed-loop guarantees?
Master's Theses | |
---|---|
Topic | Status |
Learning Linguistic Representations from iEEG Spectrogramms using Deep Learning (opens in new tab)
Co-Supervisor: Keivan Ahmadi |
available |
Enhancing EEG-based Cognitive State Classification through Graph Neural Networks (opens in new tab)
Co-Supervisor: Keivan Ahmadi |
available |
Augmenting EEG Data for Music-Emotion Recognition using Diffusion Models (opens in new tab)
Co-Supervisor: Keivan Ahmadi |
in progress |
Formation Path Planning using Model Predictive Control (opens in new tab) | in progress |
Exploring Music Perception and Imagination through Deep Learning (opens in new tab)
Co-Supervisor: Keivan Ahmadi |
in progress |
Imitative Model Predictive Control for Safe Navigation Co-Supervisor: Philipp Holzmann |
in progress |
Data-driven Surrogate Model Generation for Automated Directional Drilling (opens in new tab)
Co-Supervisor: Felix Häusser |
in progress |
Tire-Friction Learning for Vehicles by Gaussian Process State Space Models | completed |
Control of a Three-Tank System using Multi-Fidelity Gaussian Processes (opens in new tab)
Co-Supervisor: Felix Häusser |
completed |
Graph Diffusion for Imitation Learning in Robotics | completed |
Stochastic Nonlinear Model Predictive Control for Offshore Energy Systems using Gaussian Processes | completed |
Set-up and Closed-Loop Control of an Exoskeleton Co-Supervisor: Sebastian Hirt |
completed |
Learning Patient Models of Acute Lymphoblastic Leukemia for Individualized Maintenance Therapy | completed |
Modeling of Cell Dynamics During Maintenance Therapy of Acute Lymphoblastic Leukemia | completed |
Cost Function Learning for Model Predictive Control Using Bayesian Optimization | completed |
Physically Consistent Model Learning of Robotic Systems with Gaussian Processes (opens in new tab)
Co-Supervisor: Philipp Holzmann |
completed |
Gaussian-Process-based Modeling of Human Drivers from Real Data (opens in new tab)
Co-Supervisor: Johanna Bethge |
completed |
The Fokker-Planck Equation for Gaussian Process-based Model Predictive Control (opens in new tab) | completed |
Bachelor's Theses | |
Topic | Status |
Deep Learning Approaches for Classifying EEG Responses to Naturalistic Music Stimuli (opens in new tab)
Co-Supervisor: Keivan Ahmadi |
completed |
Modeling of Human-Driven Vehicles from Real Data Using Gaussian Mixture Models | completed |
Gaussian-Process-based Modeling of Individual Human-Driven Vehicles from Real Data (opens in new tab)
Co-Supervisor: Johanna Bethge |
completed |
Project Seminars | |
Topic | Status |
Contract-Based Hierarchical Control of a Mobile Ground Robot (opens in new tab)
Co-Supervisor: Lukas Theiner |
completed |
Tuning of Model Predictive Control using Bayesian Optimization (opens in new tab)
Co-Supervisor: Philipp Holzmann |
completed |
Model Identification and Control for Robotic Manipulators using Physics-Informed Neural Networks (opens in new tab)
Co-Supervisor: Philipp Holzmann |
completed |
Iterative Model Improvement Learning Control for Robotic Manipulators (opens in new tab)
Co-Supervisor: Philipp Holzmann |
completed |
Formal Verification of a Robotic Arm using Hybrid Model Checking (Otto-von-Guericke University Magdeburg) |
completed |
Technical University of Darmstadt | |
---|---|
Model Predictive Control and Machine Learning | Winter term 2021/2022, 2022/2023, 2023/2024, 2024/2025 |
Project Course: Control Theory | Summer term 2024 |
Modeling and Simulation | Summer term 2022 |
Otto-von-Guericke University Magdeburg | |
Machine Learning, Dynamical Systems, and Control | Summer term 2020, 2021 |
Introduction to Cybernetics | Winter term 2020/2021 |
2024 | |
---|---|
[13] |
M. Pfefferkorn, R. Findeisen: Probabilistically Input-to-State Stable Stochastic Model Predictive Control. Conference on Decision and Control, 2024. Accepted. |
[12] |
S. Hirt, M. Pfefferkorn, R. Findeisen: Safe and Stable Closed-Loop Learning for Neural-Network-Supported Model Predictive Control. Conference on Decision and Control, 2024. Accepted. |
[11] |
S. Hirt, M. Pfefferkorn, R. Findeisen: Stability-informed Bayesian Optimization for MPC Cost Function Learning. Conference on Nonlinear Model Predictive Control, 2024. To appear. |
[10] |
M. Pfefferkorn, V. Renganathan, R. Findeisen: Regret and Conservatism of Distributionally Robust Constrained Stochastic Model Predictive Control. American Control Conference, 2024. To appear. |
[9] |
M. Pfefferkorn, V. Renganathan, R. Findeisen: Regret and Conservatism of Distributionally Robust Constrained Stochastic Model Predictive Control. arXiv, 2309.12190, 2024. Extended version. |
[8] |
P. Holzmann, M. Pfefferkorn, J. Peters, R. Findeisen: Learning Energy-Efficient Trajectory Planning for Robotic Manipulators using Bayesian Optimization. European Control Conference, pages 1374 – 1379, 2024. |
2023 | |
[7] |
A. Rose, M. Pfefferkorn, H. H. Nguyen, R. Findeisen: Learning a Gaussian Process Approximation of a Model Predictive Controller with Gurantees. Conference on Decision and Control, pages 4094 – 4099, 2023. |
[6] |
J. Bethge, M. Pfefferkorn, A. Rose, J. Peters, R. Findeisen: Model Predictive Control with Gaussian-Process-Supported Dynamical Constraints for Autonomous Vehicles. IFAC-PapersOnline 56 (2), pages 507 – 512, 2023. |
2022 | |
[5] |
M. Pfefferkorn, P. Holzmann, J. Matschek, R. Findeisen: Safe Corridor Learning for Model Predictive Path Following Control. IFAC-PapersOnline 55 (30), pages 79 – 84, 2022. |
[4] |
H. H. Nguyen, M. Pfefferkorn, R. Findeisen: High-probability stable Gaussian-process-supported model predictive control for Lur'e systems. European Journal of Control 68, 100695, 2022. |
[3] |
P. Holzmann, J. Matschek, M. Pfefferkorn, R. Findeisen: Learning secure corridors for model predictive path following control of autonomous systems in cluttered environments. European Control Conference, pages 1772 – 1777, 2022. |
[2] |
M. Pfefferkorn, M. Maiworm, R. Findeisen: Exact Multiple-Step Predictions in Gaussian-Process-based Model Predictive Control: Observations, Possibilities, and Challenges. American Control Conference, pages 2829 – 2836, 2022. |
2020 | |
[1] |
M. Pfefferkorn, M. Maiworm, C. Wagner, F. S. Tautz, R. Findeisen: Fusing Online Gaussian-Process-Based Learning and Control for Scanning Quantum Dot Microscopy. Conference on Decision and Control, pages 5525 – 5531, 2020. |
since 06/2023 | Research assistant and Ph.D. student at the Control and Cyber-Physical Systems Laboratory (Prof. Rolf Findeisen), Technical University of Darmstadt (Germany) |
06/2020 – 05/2024 | Regular fellow of the graduate program Mathematical Complexity Reduction (DFG GRK 2297) at the Faculty of Mathematics, Otto-von-Guericke University Magdeburg |
03/2020 – 05/2024 | Research assistant and Ph.D. student at the Systems Theory and Automatic Control Laboratory (Prof. Rolf Findeisen), Otto-von-Guericke University Magdeburg (Germany) |
01/2020 | Master's degree (M.Sc.) in Biosystems Engineering from the Otto-von-Guericke University Magdeburg (Germany) |
05/2018 | Bachelor's degree (B.Sc.) in Biosystems Engineering from the Otto-von-Guericke University Magdeburg (Germany) |