How to Train Your Robot? Making robots smarter every day!
Speaker: Dr Petar Kormushev, Head of the Robot Intelligence Lab, Imperial College London
Abstract
In the near future, most households will likely have a domestic robot. But how would you teach your robot new tasks? Endowing robots with human-like abilities to perform physical tasks in a smooth and natural way has been a dream of many researchers. Ideally, robots should be able to acquire new skills through natural interaction with humans. However, acquiring new motor skills is not simple and involves various forms of learning. Some tasks can be successfully transferred to a robot using only imitation strategies. Other tasks can be learned better by the robot alone using reinforcement learning. The key to an efficient learning process lies in the interaction between imitation and self-improvement strategies. The talk will summarize the existing methods for robot learning of new motor skills. A variety of example tasks will be presented, such as learning to manipulate objects, learning efficient bipedal locomotion, whole-body motor skill learning, learning to recover from failures, etc. Throughout these examples, the important role of the learning algorithms will be highlighted, and their limitations revealed. The talk will finish with a glimpse of the cutting-edge research on deep reinforcement learning for robotics.
Biography
Dr Petar Kormushev is Head of the Robot Intelligence Lab at Imperial College London. He is also a Lecturer in Robotics and Computing at the Dyson School of Design Engineering. Dr Kormushev holds a PhD in Computational Intelligence from Tokyo Institute of Technology, a MSc degree in Artificial Intelligence and a MSc degree in Bio- and Medical Informatics. Before joining Imperial College, he was a research team leader at the Italian Institute of Technology, where he worked on the humanoid robots WALK-MAN, COMAN, and iCub. Dr Kormushev’s research focus is on machine learning algorithms and their application to autonomous robots. His long-term goal is to create robots that can incrementally learn by themselves and adapt to dynamic environments.