Real-time machine learning in ROS? (Outlier Detection)

asked 2019-03-15 08:18:11 -0500

JanKolbeint gravatar image

Hello,

I'm creating a signal processing module for a ROS project, and I'd like to do outlier detection to detect spurious sensor readings. Some of the ways I've looked at doing this is one-class SVM or isolation forests, but I'm not sure if it's possible to implement said algorithms in ROS where sensor readings are produced at a 10 Hz rate? Does ROS have any capacity to do real-time machine learning through for instance Python scikit, or is it nigh impossible?

Thanks a lot! Jan

edit retag flag offensive close merge delete

Comments

1

Does ROS have any capacity to do real-time machine learning through for instance Python scikit, or is it nigh impossible?

ROS (at its barest) provides you with a communication infrastructure. There is nothing "in it" that would make what you want to do impossible afaict.

Also:

I'm not sure if it's possible to implement said algorithms in ROS

to be blunt: you will not be implementing any algorithms "in ROS". At best you'll wrap an implementation in a (thin) ROS node that takes care of input and output of data.

gvdhoorn gravatar imagegvdhoorn ( 2019-03-17 15:19:52 -0500 )edit