Start Date

19-5-2023 3:50 PM

End Date

19-5-2023 4:00 PM

Document Type

Extended Abstract

Keywords

Deep Learning, Transfer Learning, Robust/Adaptive Control

Description

Pouring is an efficient way to transfer objects from
one container to another. This abstract summarizes a method
to accurately pour solid objects, such as ice cubes. It leverages
visual and proprioceptive feedback together with contextual
information to control the forward and backward rotation of the
pouring container. These feedback signals are fed to a recurrent
neural network that produces the control signal. The proposed
approach can achieve a human-like pouring accuracy in both a
simulation and a real setup.

DOI

https://doi.org/10.5038/RLHF4742

Download

Included in

Robotics Commons

Share

COinS
 
May 19th, 3:50 PM May 19th, 4:00 PM

Robot Learning to Pour Solid Objects Accurately

Pouring is an efficient way to transfer objects from
one container to another. This abstract summarizes a method
to accurately pour solid objects, such as ice cubes. It leverages
visual and proprioceptive feedback together with contextual
information to control the forward and backward rotation of the
pouring container. These feedback signals are fed to a recurrent
neural network that produces the control signal. The proposed
approach can achieve a human-like pouring accuracy in both a
simulation and a real setup.