From Brown University Robotics

Jump to: navigation, search




This project exists to turn Sparse Online Gaussian Processes (SOGP) into a ROS node, in order to provide fast approximate Gaussian Process regression in ROS. The basis of this work is Dan Grollman's implementation of Csato's thesis on SOGP. Our hope is that ROS users will use this code to do robot learning from demonstration or other types of function regression.

Point of Contact


Alex Tarvo

Jesse Butterfield

Dan Grollman

Svn Pointer

HowTo Link

Internal:SOGP Instructions


Alex Tarvo did the original documentation. Jesse has cleaned up the code, added compatibility with rosparam, and included the ability to save and load learned controllers. However, the codes need significantly better documentation and improved exception handling. Also eventually, we would like to be able to automatically detect the dimensionality of the input and output and provide a measure of the confidence.

This page was last modified on 3 June 2010, at 20:45. This page has been accessed 289 times.