From imitation to self-exploration: Learning object manipulation skills

Speaker: Seungsu Kim

Email: seungsu.kim(at)naverlabs.com

Robotics Seminar Series. Second Session – 1sh April 2022, 15:00-16:00 (EEST),  via zoom (use this link): https://aalto.zoom.us/j/62124942899

Abstract: Robots are considered as one of the solutions for automating human physical labor in the future aging society. Repetitive or tedious tasks such as cleaning up the dirty dining table after dinner or tidying up the children’s room cluttered with toys can be done by robots, and people can spend their time on things that are more valuable to them.

The robot’s decision-making system refers to a system that decides how to control the robot’s actuator to perform a given task based on the information received from various sensors. The purpose is to make the robot to acquire robust and adaptive skills so as to complete assigned tasks while adapting well to changes in the surrounding environment. Humans gradually develop these adaptive skills based on the experiences they have accumulated since birth. On the other hand, decision-making systems of a robot manipulator are still clumsy in terms of adaptability for everyday tasks.

In this talk, I will introduce various research activities for the purpose of learning robot manipulation skills, such as learning from demonstration, motor-babbling, goal-babbling, quality-diversity search and self-exploration.