ロボット学習分野におけるトップ学会の一つであるConference on Robot Learning (CoRL)に当センターの長隆之准教授が招待され、チュートリアル講演を行いました。
学会HPへのリンクはこちら
動画はこちら
講演題目： Dealing with the objective function with multiple extrema in robot learning
概要: In the real world, there are often diverse solutions to a problem. For example, when walking at a certain speed, there can be infinitely many walking styles to achieve the specified speed. The existence of multiple solutions is often originated from the multiple extrema in an objective function in robot learning. While discovering a single solution may be sufficient to solve the problem at hand, extracting diverse behaviors from a single task could be an efficient way to obtain skill repertoires for downstream tasks. In this talk, we will discuss methods for discovering diverse solutions in the context of motion planning and reinforcement learning. We will first discuss the problem formulation for discovering diverse solutions in motion planning and how to solve it. Next, we will discuss reinforcement learning methods for discovering diverse policies. This talk aims to provide basics of discovering diverse solutions in robot learning and share our experience through case studies.
