CSS437 Introduction to Robotics

Control Theory

Cybernetics - the science of control in systems.  Norbert Weiner.

Stability

A controlled process maintains output values within critical limits
External influences disturb the stability of the process
 

Open-loop control

When no outside influences can disturb the functioning of a process.
Not very likely in the real world of mobile, autonomous robots!!!

Principle of feedback - Closed-loop control in the "real" world

Feedback works by counteracting the effects of a disturbance or error in desired vs. actual.
Measuring a critical output parameter
Comparing it to an ideal value (setpoint)
Using error to adjust the process
Error has both direction and magnatude
 

Proportional Control

In proportional control, the control signal, C, is proportional to the direction and magnatude of the error signal.
C = k0 e
where k0 is a gain constant (note that it is posible to introduce more complex functions including non-linear ones but this is problematic in most cases).
 
Problems
Overcompensation and oscillations.  Inertia. Non-linearity.

Derivative Control

In derivative control, the control signal is proportional to the rate of change in the error signal (not a pure control strategy). Used to adjust the overall gain and damp oscillations.
C = k1 de/dt

Integrative Control

In integrative control, the control signal is proportional to the integral of error over some time frame.  Ususally implelented as a moving average. (Also not a pure control strategy).  Used to smooth the response of a control signal.
C = k2 I e(t) dt
 

PID Control

Combining these three sources of information into a single control signal provides the greatest stability in systems with high degrees of disturbance and control inertia.
C = k0 e + k1 de/dt + k2 I e(t) dt

The relative influences of error, rate of change in error and average error over time are controlled by the gain constants, k0, k1 and k2.  Combinational controls such as PD or PI are possible.
 

Feedforward Control - Anticipation

The purpose of feedforward control is to take preemptive action in order to further reduce the effects of oscillation or lag in adjusting the control.  It assumes that a disturbance will change the values of inputs from the ideal.  Adjustments to the control signal can be made before a noticable effect is measured at the output.
 
 

Hierarchical Control

A more sophisticated form of control.  This form assumes that there are multiple inputs and outputs which can be varied to produce an optimal mix of desired outputs.

Adaptive Control

The most sophisticated controllers combine hierarchical and feedforward mechanisms in multiple layers.  Adaptive controllers include memory and models of the environment to further refine anticipatory action.  Models of the environment are modifiable over time to compensate for changes in the environment.