Public defence in Computer Science, MSc Amin Babadi

Title of the doctoral thesis is "Advances in Optimizing Physically Simulated Movements"
In the presence of a high-quality reference trajecotry, the sample efficiency of the optimizer could be increased by limiting the maximum amount of allowed deviation from that trajectory.

Amin Babadi's doctoral dissertation advances the movement optimization of physically simulated characters by proposing and evaluating several optimization, learning, and visualization approaches. Physically simulated movement optimization is a complex problem with applications in, e.g., robotics and games. Over the years, several approaches have been proposed, but the field still suffers from either high computational cost, or the dependency on reference animation data that may be slow, expensive, and hard to obtain.

The dissertation focuses on the movement optimization problem by exploring two popular approaches: trajectory optimization and reinforcement learning (RL). A key advantage of the methods proposed in this dissertation is that they are general and can be applied to various physically simulated characters without having to limit oneself to characters for which reference movement data is available. The methods are suitable for time-sensitive settings such as games. Finally, the dissertation contributes by proposing novel visualizations of the movement optimization problem and how various problem modifications affect the optimization landscape.

Opponent: Doctor Daniel Holden, Epic Games, Canada

Custos: Professor Perttu Hämäläinen, Aalto University School of Science, Department of Computer Science

Contact details of the doctoral student: [email protected]

The public defence will be organised on campus (R001/H304, Otakaari 1).

The thesis is publicly displayed 10 days before the defence in the publication archive Aaltodoc of Aalto University. 

Electronic thesis

  • Published:
  • Updated: