Difference between revisions of "Is there room for learning in control systems theory?"

From MurrayWiki
Jump to: navigation, search
Line 1: Line 1:
 
In the "sensing," "computation," and "actuation" blocks of a control system, learning happens in the computation block. Broadly speaking learning involves techniques for exploring the parameter space and measuring the system's performance relative to a set of criteria. For example, one might use a learning controller to optimize the gains in a PID controller (proportional-integral-derivative, this is a standard controller type we'll learn more about later in the course). We won't be covering machine learning in this course, but it is an interesting area in controls.  
 
In the "sensing," "computation," and "actuation" blocks of a control system, learning happens in the computation block. Broadly speaking learning involves techniques for exploring the parameter space and measuring the system's performance relative to a set of criteria. For example, one might use a learning controller to optimize the gains in a PID controller (proportional-integral-derivative, this is a standard controller type we'll learn more about later in the course). We won't be covering machine learning in this course, but it is an interesting area in controls.  
  
I'll try to post a reference on learning in the next few days.
+
from Richard:
 +
 
 +
* In robotics, there is a large literature on iterative learning
 +
control, when you are performing a repetitive task.  This mainly
 +
involves turning the trajectory generation (feedforward).
 +
 
 +
* Adaptive control can be considered a type of learning (based on
 +
adjusting either a model of the plant or the parameters of a control).
  
 
--[[User:Fuller|Sawyer Fuller]] 00:21, 2 October 2007 (PDT)
 
--[[User:Fuller|Sawyer Fuller]] 00:21, 2 October 2007 (PDT)

Revision as of 22:50, 8 October 2007

In the "sensing," "computation," and "actuation" blocks of a control system, learning happens in the computation block. Broadly speaking learning involves techniques for exploring the parameter space and measuring the system's performance relative to a set of criteria. For example, one might use a learning controller to optimize the gains in a PID controller (proportional-integral-derivative, this is a standard controller type we'll learn more about later in the course). We won't be covering machine learning in this course, but it is an interesting area in controls.

from Richard:

  • In robotics, there is a large literature on iterative learning

control, when you are performing a repetitive task. This mainly involves turning the trajectory generation (feedforward).

  • Adaptive control can be considered a type of learning (based on

adjusting either a model of the plant or the parameters of a control).

--Sawyer Fuller 00:21, 2 October 2007 (PDT)