22 May - 2018 (Last edit: 24 May - 2018)
NWO awards TOP-grant to Guido de Croon for self-learning robots
The Netherlands Organisation for Scientific Research (NWO) has awarded a TOP-grant, a personal grant consisting of a funded research position, to Guido de Croon for his work on self-learning robots. The grant aims to fund “innovative or high-risk scientific research that addresses questions of high quality and urgency withrelevance for astronomy, computer science, and/or mathematics.” De Croon obtained a Module 2 grant, which is available for researchers who have received their PhD within the previous ten years and aims to assist the researcher in setting up a research group. De Croon plans to use the grant for research into improving the abilities of robots to learn by themselves how to perceive their environment.
Autonomous robots are going to profoundly change our society and industry, where self-driving cars and package delivery drones are only the tip of the iceberg. It is commonly believed that robots will be able to perform their tasks better if they are able to learn from their environment. However, most robotic learning mechanisms are difficult to port to the real world. Notably, the very successful deep supervised learning methods can be difficult to generalize to a robot’s actual environment, as the environments in the training set may be too different.
Self-supervised learning (SSL)
Guido de Croon proposes self-supervised learning (SSL) as a reliable mechanism to have robots learn in their own environment. In SSL, robots have a baseline perception and behavior module, which allows for initial operation while in the same time providing supervised targets to a learning process that extends their capabilities. Where previous work has shown the potential of SSL in isolated case studies, De Croon will forward the first structured study of SSL. The key innovation is the extension of SSL with novel elements and identify and propose solutions to fundamental challenges in the way of its widespread usage. In particular, De Croon will study the guarantee of a baseline performance, the fusion of baseline and learned perceptual capabilities, coping with “bad learning” by means of introspection, and active and collaborative learning.
“In the research I will uncover the full potential of SSL, both theoretically and empirically” says Guido de Croon. “This will make SSL a learning mechanism of choice, with a potential impact on any autonomous robot.”
NWO release (Dutch) “5,7 miljoen euro voor toponderzoek in astronomie, informatica en wiskunde” https://www.nwo.nl/actueel/nieuws/2018/05/57-miljo...