Autonomy Talks - 18/10/22
Speaker: Annie Xie, Stanford University
Title: Towards Continual Robotic Reinforcement Learning
Abstract: Humans are constantly learning about their surroundings and how to carry out new tasks in them. In contrast, typical robotic learning set-ups are rigidly segmented into phases of training followed by deployment. This talk focuses on how we can move towards continual learning systems that seamlessly and, more importantly, autonomously acquire new skills under the reinforcement learning framework. I will cover recent work that enables robotic systems to quickly learn new tasks through data reuse. Then, I will discuss how we can overcome conditions of the physical world, such as non-stationarity and irreversibility, that make lifelong learning particularly challenging.
If you want to see more, please visit [ Ссылка ]
