March 16, 2026
The Institute of Informatics & Telecommunications will host invited speaker George Kamaras for a talk titled Real2Sim2Real for Vision-Driven Deformable Linear Object Manipulation on Thursday 2 April 2026 at 11.00am EET.
The talk will be held in person at the Aegeo meeting room of the Institute of Informatics & Telecommunications.
About the talk: Humans excel in manipulating deformable linear objects (DLOs), such as belts, cables, and shoelaces in various everyday tasks. Yet for robotics, visuomotor DLO control is an open problem due to the high-dimensional nonlinear dynamics and the stochasticity involved in state estimation. We can train an agent in simulation to learn a successful policy for such tasks while minimising costly real-world experiments. However, a simulated environment can vary so significantly from its real counterpart that a policy trained in simulation may fail when deployed in a physical setting that appears qualitatively close. To overcome these challenges, we propose a vision-driven Real2Sim2Real framework, which bridges the ideas of likelihood-free inference and domain randomisation and enables object-centric agent skill development.
Short Bio: Georgios Kamaras is a PhD student (viva passed; pending award) in Robotics and Autonomous Systems at the University of Edinburgh, supervised by Prof. Subramanian Ramamoorthy and Dr. Kartic Subr. He is a member of the Centre for AI in Assistive Autonomy and his research has been supported by EPSRC through the Edinburgh Centre for Robotics. His doctoral work focuses on advancing deformable object manipulation through reinforcement learning, integrating likelihood-free inference and domain randomisation to address the simulation-to-reality gap. He holds a BSc in Informatics and Telecommunications from the National and Kapodistrian University of Athens, and is an alumnus of the Roboskel group at NCSR Demokritos, where he worked on human-robot interaction and path planning for incline terrain navigation.