LEPCO-EX2 - Learning-based Model Predictive Control by Exploration and Exploitation in Uncertain Environments

Start

01/12/2023

End

31/03/2026

Status

In progress

LEPCO-EX2 - Learning-based Model Predictive Control by Exploration and Exploitation in Uncertain Environments

This project develops a framework for the simultaneous learning and control of dynamic systems that can cope with uncertain and time-varying dynamics or environments within a receding horizon control framework. The proposed strategies will integrate active learning actions to exploit information about the process generated online during optimal closed-loop operation, without artificially generating informative conditions and without performing independent or enforced system identification experiments during operation.

One of the considered applications is Active Noise Control (ANC) in time-varying environments. In this setting, starting from a partially known model of the environment, the control policy must learn the wave propagation model of changing spaces by trading off noise cancellation and the generation of information about system behavior.

Considering the difficulty of measuring acoustic pressure in advance to apply classical ANC methods, artificial intelligence tools such as deep learning algorithms can provide an effective data-driven approach. They can be applied to compute/predict the acoustic pressure due to their ability to capture spatial coherent patterns in the radiated acoustic pressure fields. More specifically, this project develops real-time implementations of model predictive control (MPC) algorithms based on deep learning on dedicated modular hardware at the PoliMi Sound and Vibration Laboratory.

Publications

Y. He, H.R. Karimi, P. Mei, Cooperative multi-agent deep reinforcement learning-based eco-driving strategy for hybrid electric vehicles at multi-intersection scenarios, Neurocomputing, 132412, (2025)

L.J. Liu, T.T. Huang, H.R. Karimi, Y.H. Ma, J. Su, Optimized traffic signal control system incorporating mixed traffic flow and adverse weather, Applied Intelligence 55 (17), 1115, (2025)

Y. Yu, H.R. Karimi, L. Gelman, J. Tian, P. Mei, A novel multi-source sensor correlation adaptive fusion framework with uncertainty quantification for intelligent fault diagnosis, Reliability Engineering & System Safety, 111812, (2025)