From Supervised to Reinforcement Learning: an Inverse Optimization Approach

From Supervised to Reinforcement Learning: an Inverse Optimization Approach

Dimanidis, Ioannis (TU Delft Mechanical, Maritime and Materials Engineering)

Mohajerin Esfahani, P. (mentor)
Mazo, M. (graduation committee)
Atasoy, B. (graduation committee)

Delft University of Technology

Mechanical Engineering | Systems and Control

2021-12-10

We propose a novel method combining elements of supervised- and Q-learning for the control of dynamical systems subject to unknown disturbances. By using the Inverse Optimization framework and in-hindsight information we can derive a causal parametric optimization policy that approximates a non-causal MPC expert. Furthermore, we propose a new min-max MPC scheme that robustifies against a ball around a disturbance trajectory. This scheme yields an exact convex reformulation using the S-Lemma, and is also approximated using Inverse Optimization. Finally, simulation studies clarify and verify our approach.

Inverse Optimization
Reinforcement Learning
Convex Optimization
Optimal Control
data-driven control

http://resolver.tudelft.nl/uuid:9d5efafe-58e5-497a-b7ea-0ac0ce2e9173

Student theses

master thesis

© 2021 Ioannis Dimanidis