Canada Spends Logo

Explorez notre analyse du budget 2026

Grant: Discovery Grants Program - Individual

The University of British Columbia

$32,000.00

Data-driven modeling and control for safety-critical applications

EF 2024-25

Résumé
Advances in sensing, data accessibility and machine learning have hugely increased interest in learning control, i.e. methods that learn from data to design or improve control systems. Reinforcement learning in robotics, data-driven control for complex power networks and advances in autonomous driving have moved learning control into applications with widespread societal impact. Since many new learning methods do not translate well from simulation to the real world, the importance of "Safety-critical autonomous systems with Machine Learning" as featured in the 2021 IEEE CSS workshop on Control for Societal Challenges, is well recognized in the control systems community. Learning from data is advantageous when accurate models are not available or expensive to develop, but state-of-the-art safety-enforcement algorithms often rely on accurate models, placing them at odds with the learning control they aim to enhance. My long-term vision is to bridge the gap between the success of learning control in simulation and its performance in real-world applications, through development of safe and data-efficient algorithms that exploit ideas from `identification for control'. I aim to redefine the notion of accurate models, by explicitly linking the required information (model accuracy) to the learning objective (e.g. safety, stability, performance). This DG program is motivated by my work on highly uncertain safety-critical medical control systems, for which accurate models are not available. For such systems, solutions using cutting-edge modeling and safety-enforcement algorithms are still so conservative that they do not allow for adequate control. Recent literature, including my work extending the current framework, emphasizes that the commonly used state-space formulation is highly limiting for systems with unmodeled dynamics. In this DG program, I propose a new approach to address shortcomings of current safety-critical control solutions through more appropriate modeling that captures the information required to ensure safety. I will develop new methods to improve performance of safe control through the use of two estimates, one for performance and one that better predicts unsafe situations. In addition, I propose a novel data-driven safe control formulation that is simpler and requires fewer resources to compute. In this DG program, five HQP will receive training in advanced control, modeling and learning control, and gain skills that are highly desired in industry and academia. These HQP will have thorough understanding of the opportunities and challenges in ensuring the autonomous systems of our future are safe. Through the development of innovative methods to improve safety-critical control, this research will expand the state-of-the-art, offer opportunities for follow up research and provide important advances to extend the applicability of methodology for safe autonomous systems from toy/trivial examples to real-world applications.
Programme
Discovery Grants Program - Individual
Emplacement
British Columbia, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data

Tel que publié

Application id
RGPIN-2023-03660
Name nom
van Heusden, Klaske
Department département
Okanagan - School of Engineering
Organization id
2
Institution établissement
The University of British Columbia
Province en
British Columbia
Province fr
Colombie-Britannique
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2023
Award amount
32000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1512
Committee name en
Mechanical Engineering
Committee name fr
Génie mécanique
Area of application code
1206
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Engineering
Area of application fr
Génie
Research subject code
2509
Research subject group en
Electrical and electronic engineering
Research subject group fr
Génie électrique et électronique
Research subject en
Control systems
Research subject fr
Asservissements
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Data-driven modeling and control for safety-critical applications
Keywords
control systems; data-driven control; learning control; system identification; safety-critical control systems; constrained control
Application summary
Advances in sensing, data accessibility and machine learning have hugely increased interest in learning control, i.e. methods that learn from data to design or improve control systems. Reinforcement learning in robotics, data-driven control for complex power networks and advances in autonomous driving have moved learning control into applications with widespread societal impact. Since many new learning methods do not translate well from simulation to the real world, the importance of "Safety-critical autonomous systems with Machine Learning" as featured in the 2021 IEEE CSS workshop on Control for Societal Challenges, is well recognized in the control systems community. Learning from data is advantageous when accurate models are not available or expensive to develop, but state-of-the-art safety-enforcement algorithms often rely on accurate models, placing them at odds with the learning control they aim to enhance. My long-term vision is to bridge the gap between the success of learning control in simulation and its performance in real-world applications, through development of safe and data-efficient algorithms that exploit ideas from `identification for control'. I aim to redefine the notion of accurate models, by explicitly linking the required information (model accuracy) to the learning objective (e.g. safety, stability, performance). This DG program is motivated by my work on highly uncertain safety-critical medical control systems, for which accurate models are not available. For such systems, solutions using cutting-edge modeling and safety-enforcement algorithms are still so conservative that they do not allow for adequate control. Recent literature, including my work extending the current framework, emphasizes that the commonly used state-space formulation is highly limiting for systems with unmodeled dynamics. In this DG program, I propose a new approach to address shortcomings of current safety-critical control solutions through more appropriate modeling that captures the information required to ensure safety. I will develop new methods to improve performance of safe control through the use of two estimates, one for performance and one that better predicts unsafe situations. In addition, I propose a novel data-driven safe control formulation that is simpler and requires fewer resources to compute. In this DG program, five HQP will receive training in advanced control, modeling and learning control, and gain skills that are highly desired in industry and academia. These HQP will have thorough understanding of the opportunities and challenges in ensuring the autonomous systems of our future are safe. Through the development of innovative methods to improve safety-critical control, this research will expand the state-of-the-art, offer opportunities for follow up research and provide important advances to extend the applicability of methodology for safe autonomous systems from toy/trivial examples to real-world applications.