Differentiable Programming for Modeling and Control of Dynamical Systems

Explore differentiable programming for optimal control of dynamical systems in this insightful talk. Delve into differentiable predictive control (DPC) as a model-based policy optimization method, integrating classical model predictive control principles with scientific machine learning. Learn how domain-aware differentiable models for dynamical systems are leveraged for offline policy optimization under nonlinear constraints. Discover DPC's scalability, data efficiency, and constraints handling through simulation case studies in building control and dynamic economic dispatch. Examine the experimental demonstration of DPC's computational and memory efficiency in embedded implementation on a laboratory device, showcasing its potential for control as a service applications.