Shahbaz Abdul Khader
I am a PhD student at the Robotics, Perception, and Learning Lab, KTH under the supervision of Prof. Danica Kragic. I am also employed at ABB Corporate Research, Västerås, Sweden as a Scientist. My work is supported by the Wallenberg AI, Autonomous Systems and Software Program (WASP) funded by the Knut and Alice Wallenberg Foundation.
As an enthusiast of autonomous robots, I am interested in applying machine learning for solving control and decision making problems. My research focuses on how robots can autonomously acquire complex manipulation skills in uncertain and unstructured environments.
While manipulating in uncertain and unstructured environments, physical contact between the manipulator and the environment is almost a certainty. Since traditional planning and control methods rarely scale for contact-rich manipulation, I adopt a skill learning approach where a robot can autonomously acquire rich control policies through reinforcement learning (RL). The learned policies can not only generate motion but also the right interaction or compliance behavior.
In my work, I explore two of the most important aspects in RL-based skill learning: control stability and data-efficiency. Control stability addresses the question of how to ensure stability, in the sense of Lyapunov, during the RL process. Data-efficiency addresses the need for sample efficient RL. These two aspects are studied in the context of contact-rich manipulation. See the portfolio pages for more information.