Events Calendar

Reinforcement Learning: Towards Optimal Adaptive Control

This is a past event.

Monday, March 25, 2019 4 pm

This is a past event.

ME-EM Research Seminar Speaker Series

proudly presents:

Dr. Enrico Anderlini

University College London, UK

Abstract: Control systems regulate the behavior of devices. Starting with the description of some control strategies, optimal control is shown to provide optimal performance when accounting for external disturbances, while adaptive control is important to deal with varying system dynamics, e.g. due to components’ failures or marine growth. An alternative approach to standard control schemes is represented by reinforcement learning, which provides a framework where the controller learns the optimal behavior from direct interactions with the environment. The method is also adaptive and nonlinear, since it does not rely on a model of the system dynamics. Reinforcement learning is explained in the presentation through the application to the control of a wave energy converter. However, the described implementation relies on the analysis of time-averaged wave conditions. To achieve a step change in performance required to reduce the levelized cost of wave energy, a real-time control solution is required instead. This is now possible thanks to the latest advances in reinforcement learning algorithms, with function approximation for both the state and action spaces being provided by deep learning. A further drop in learning time can be achieved through apprenticeship learning by learning from expert operators if the information is available, as is the case with the control of underwater vehicles. Finally, an outlook to the expected future direction of this research, including funding, is provided.

Bio: Enrico Anderlini is a research associate in naval architecture at University College London (UCL). His research focuses on the control of unoccupied underwater vehicles (UUVs) and wave energy converters (WECs), with an interest in reinforcement learning. As a result, he has been contributing to the refurbishment and validation studies for the UCL wave tank and the acquisition and development of four UUVs. Additionally, he set up a collaboration scheme with the National Oceanography Centre in Southampton, in which he has developed tools for the automation of their fleet of underwater gliders. Previously, he completed his engineering doctorate on the application of reinforcement learning to the control of WECs at the Industrial Doctoral Centre for Offshore Renewable Energy under the sponsorship of Wave Energy Scotland.

Additional Details


  • Sal Husain

1 person added

User Activity

No recent activity