Colin Jones

Colin Jones
Prof.
Colin Jones
PI
Executive Committee Member
Control is full of fascinating mathematical challenges that can make a real difference to the world, especially the green energy transition.

Colin Jones has been an Associate Professor in the Automatic Control Laboratory at the EPFL in Switzerland since 2017 and an assistant professor from 2011. He was a Senior Researcher at the Automatic Control Lab at ETH Zurich until 2010 and obtained a Ph.D. in 2005 from the University of Cambridge for his work on polyhedral computational methods for constrained control. He is the author or coauthor of more than 200 publications and was awarded an ERC starting grant to study the optimal control of building networks. His current research interests lie at the intersection of machine learning, predictive control and optimization, as well as the control of green energy generation, distribution and management.

Scientific Publications

Published
A Generalized Stopping Criterion for Real-Time MPC with Guaranteed Stability
2023 62nd IEEE Conference on Decision and Control (CDC)
Published
Stable Linear Subspace Identification: A Machine Learning Approach
ArXiv
Published
Physically Consistent Multiple-Step Data-Driven Predictions Using Physics-Based Filters
IEEE Contro Systems Letters
Vol 7 Pages 1885-1890
Published
Relaxed Recentered Log-Barrier Function based Nonlinear Model Predictive Control
ECC 2023
Published
Adaptive Data-Driven Predictive Control as a Module in Building Control Hierarchy: A Case Study of Demand Response in Switzerland
ArXiv
Published
Hypergraph Based Fast Distributed AC Power Flow Optimization
ArXiv
Published
Physics-Informed Machine Learning for Modeling and Control of Dynamical Systems
2023 American Control Conference (ACC)
Pages 3735-3750
Published
Relaxed Recentered Log-Barrier Function based Nonlinear Model Predictive Control
2023 European Control Conference (ECC)
Pages 1-6
Published
Primal-Dual Contextual Bayesian Optimization for Control System Online Optimization With Time-Average Constraints
ArXiv
Published
PIQP: A Proximal Interior-Point Quadratic Programming Solver
62nd IEEE Conference on Decision and Control
Published
Adaptive Robust Data-Driven Building Control via Bilevel Reformulation: An Experimental Result
IEEE Transactions on Control Systems Technology
Vol 31 No 6 Pages 2420 - 2436
Published
Optimal Thrust Vector Control of an Electric Small-Scale Rocket Prototype
2022 International Conference on Robotics and Automation (ICRA)
Pages 1996-2002
Published
Computationally Efficient Reinforcement Learning: Targeted Exploration leveraging simple Rules
ArXiv
Published
CONFIG: Constrained Efficient Global Optimization for Closed-Loop Control System Optimization With Unmodeled Constraints
ArXiv
Published
Physically Consistent Neural ODEs for Learning Multi-Physics Systems
Proceedings of IFAC World congress 2023
to Appear
Physically Consistent Neural Networks for building thermal modeling: theory and analysis
Applied Energy
Vol 325
Published
Constrained Efficient Global Optimization of Expensive Black-Box Functions
ArXiv
Published
Distributed data-driven predictive control for cooperatively smoothing mixed traffic flow
Transportation Research Part C: Emerging Technologies
Vol 155
to Appear
VABO: Violation-Aware Bayesian Optimization for Closed-Loop Control Performance Optimization with Unmodeled Constraints
arxiv
Published
Near-optimal Deep Reinforcement Learning Policies from Data for Zone Temperature Control
IEEE 17th International Conference on Control & Automation (ICCA)
Pages 698-703
Published
Stability Verification of Neural Network Controllers using Mixed-Integer Programming
IEEE Transactions on Automatic Control
Published
Data-driven input reconstruction and experimental validation
arXiv
Published
Robust Resource-Aware Self-Triggered Model Predictive Control
IEEE Control Systems Letters
Pages 1724-1729
Published
Joint Energy Management for Distributed Energy Harvesting Systems
SenSys '21: Proceedings of the 19th ACM Conference on Embedded Networked Sensor Systems
Pages 575-577
Published
Deep Reinforcement Learning for Room Temperature Control: A Black-box Pipeline from Data to Policies
CISBAT 2021 special issue of IOP's JOURNAL OF PHYSICS Conference Series
Vol 2042
Published
Resource-Aware Stochastic Self-Triggered Model Predictive Control
IEEE Control Systems Letters
Vol 6 Pages 1262 – 1267

Research projects

Title
Principal Investigators

Deep Reinforcement Learning for Building Control: Physics-inspired Methods

Summary

Buildings consume 30% and 40% of the end-use energy in Europe and worldwide respectively, but the increasing amount of sensors installed in new or retrofitted buildings gives rise to new data-driven control opportunities. Most controllers in existing buildings are however still rule-based, due to the required engineering to develop more advanced methods, such as Model Predictive Control (MPC). On the other hand, Reinforcement Learning (RL) has received increasing attention in the past years as a decision making paradigm that bypasses the need for models. Instead, an agent learns to take optimal decisions solely by interacting with the environment and getting rewarded/penalized for it.

We conjecture that incorporating prior physics-based knowledge in the model architectures or learning procedures will be a stepping stone towards generally applicable - i.e. data-efficient and reliable - DRL approaches. In this work, we will thus explore various solutions to introduce prior knowledge in models and DRL agents, design algorithms to apply them to buildings, and deploy the resulting controllers on case studies to assess their advantages in practice, primarily focusing on the building energy systems installed at the Empa NEST demonstrator.

Deep Reinforcement Learning for Building Control: Physics-inspired Methods

Buildings consume 30% and 40% of the end-use energy in Europe and worldwide respectively, but the increasing amount of sensors installed in new or retrofitted buildings gives rise to new data-driven control opportunities. Most controllers in existing buildings are however still rule-based, due to the required engineering to develop more advanced methods, such as Model Predictive Control (MPC). On the other hand, Reinforcement Learning (RL) has received increasing attention in the past years as a decision making paradigm that bypasses the need for models. Instead, an agent learns to take optimal decisions solely by interacting with the environment and getting rewarded/penalized for it.

We conjecture that incorporating prior physics-based knowledge in the model architectures or learning procedures will be a stepping stone towards generally applicable - i.e. data-efficient and reliable - DRL approaches. In this work, we will thus explore various solutions to introduce prior knowledge in models and DRL agents, design algorithms to apply them to buildings, and deploy the resulting controllers on case studies to assess their advantages in practice, primarily focusing on the building energy systems installed at the Empa NEST demonstrator.

189
077ed802-beee-4804-94de-2b0e3109b69d

Dependable Distributed and Hierarchical Control under Energy Constraints

Summary

We will investigate the theoretical and practical challenges of using energy harvesting to power nodes distributed control systems. Combining energy sources such as temperature and vibration with battery systems and wireless links enables the placement of sensor nodes where they are needed for the best data quality, regardless of the availability of wired power or communication. In the context of the NCCR, we will jointly investigate some of the open challenges in the design of autonomous, energy-neutral automation systems. In a second phase, we will investigate suitable demonstrators and applications such as wireless sensing and control in motor control and energy systems.

Dependable Distributed and Hierarchical Control under Energy Constraints

We will investigate the theoretical and practical challenges of using energy harvesting to power nodes distributed control systems. Combining energy sources such as temperature and vibration with battery systems and wireless links enables the placement of sensor nodes where they are needed for the best data quality, regardless of the availability of wired power or communication. In the context of the NCCR, we will jointly investigate some of the open challenges in the design of autonomous, energy-neutral automation systems. In a second phase, we will investigate suitable demonstrators and applications such as wireless sensing and control in motor control and energy systems.

116
64602cdc-7659-4ea1-a39f-1ded30d42b54

Learning Fast Convex Optimizers

Summary

This project will study formal methods for reduced complexity design and verification of embedded optimization techniques for the control of high-speed nonlinear constrained systems. We will focus on machine learning techniques for differentiable parametric optimization and their application to the control of fast, nonlinear dynamic systems with a power conversion system taken as an important exemplar case study.

Learning Fast Convex Optimizers

This project will study formal methods for reduced complexity design and verification of embedded optimization techniques for the control of high-speed nonlinear constrained systems. We will focus on machine learning techniques for differentiable parametric optimization and their application to the control of fast, nonlinear dynamic systems with a power conversion system taken as an important exemplar case study.

114
8a64bfd0-5a4a-4ae8-951a-bad586329d22

Predictive Control for Additive Manufacturing

Summary

Selective laser melting is an additive manufacturing technology that uses a laser to melt powdered metal into a solid structure based on a sliced 3D CAD model. Currently, such processes are only controlled by setting the process parameters and set points in advance, following multiple process optimisation trials. In this project, process parameters will be optimized with a data-driven closed loop control approach, based on combination of predictive and iterative learning control.

Predictive Control for Additive Manufacturing

Selective laser melting is an additive manufacturing technology that uses a laser to melt powdered metal into a solid structure based on a sliced 3D CAD model. Currently, such processes are only controlled by setting the process parameters and set points in advance, following multiple process optimisation trials. In this project, process parameters will be optimized with a data-driven closed loop control approach, based on combination of predictive and iterative learning control.

108
4cf5b44b-ef93-4b8a-bf72-5400eacf6831