Self-learning algorithms for magnetic manipulation

In his work, Aram Simonian expanded the control system for magnetic manipulation of Magman so that the controller itself adapted its behavior according to the platforms. Aram, under the leadership of Jiri Zemanek, spent time studying several self-learning algorithms from the field of Reinforcement Learning (RL), namely: Integral Reinforcement Learning (IRL), Least-Squares-Policy-Iteration (LSPI) and Energy-Balancing-Actor-Critic Control (EBAC). These algorithms were not only theoretically analyzed, but also tested in simulations as well as in a real system. Aram not only solved complicated theoretical issues, but also practical aspects related to the implementation of the control system on the PC and the special platform, Speedgoat. During his work, Aram was in close contact with experts from TU Delft and colleagues of prof. Lewis of the University of Texas in Arlington. His work was awarded the Dean’s prize.