EE474 Robotics and Control

Lecture 5: Motor Control

Dr Gordon Dobie, EEE

 Motor Control
• Early mechanical examples
• Feed-forward and Feedback control
• Terminology Lecture 4:
• Basic controllers: Motors

– Feed-Forward (FF) control

– Bang-Bang control
Lecture 5:
– Proportional (P) control Motor Control
– The D term: Proportional-Derivative
(PD) control
– The I term: Proportional-Integral (PI)
Lecture 6:
control Motor Control
– Proportional-Integral-Derivative (PID)
control
• Gain selection
• Applications
*Material adapted from
http://www.roboticscourseware.org/
 The Role of Control
• Many tasks in robotics are defined by
achievement goals
– Go to the end of the maze
– Push that box over here
• Other tasks in robotics are defined by
maintenance goals:
– Drive at 0.5m/s
– Balance on one leg
 The Role of Control
• Control theory is generally used for low-level
maintenance goals
• General notions:
– output = Controller(input)
– output is control signal to actuator (e.g., motor
voltage/current)
– input is either goal state or goal state error (e.g.,
desired motor velocity)
• Controller is stateless
What is the point of control?
• Consider any mechanism with adjustable DOFs*
(e.g. a valve, furnace, engine, car, robot…)

• Control is purposeful variation of these DOFs

to achieve some specified maintenance state
– Early mechanical examples: float valve, steam governor

www.freshwatersystems.com

*DOFs = Degrees of Freedom

 Motor Control: Open Loop
• Give robot task with no concern for the
environment
• Applications: ???
• Open loop: signal to action
• Not checked if correct action was taken
• Example: go forward for 15 secs, then turn
left for 10 secs. Issues?

signal actuator output

 Open loop (feed-forward) control
• Open loop controller:
– output = FF(goal)
• Example.: motor speed controller (linear):
– V=k *s
– V is applied voltage on motor
– s is speed
– k is gain term (from calibration)
• Weakness:
– Varying load on motor => motor may not maintain goal
speed
Feed-Forward (FF) Control
• Pass command signal from external environment
directly to the loaded element (e.g., the motor)
• Command signal typically multiplied by a gain K
Command
signal
 Motor RPM

K
• … Where does the gain value K come from?

• Under what conditions will FF control work well?

Feed-Forward (FF) Control
• Pass command signal from external environment
directly to the loaded element (e.g., the motor)
• Command signal typically multiplied by a gain K
Command
signal
 Motor RPM

K
• … Where does the gain value K come from?
– Calibration (example: PWM = 0, PWM = 255)
• Under what conditions will FF control work well?
– When the presented load is uniform and known
 Feedback Control
• Feedback controller:
– output = FB(error)
– error = goal state - measured state
– controller attempts to minimise error
• Feedback control requires sensors:
– Binary (at goal/not at goal)
– Direction (less than/greater than)
– Magnitude (very bad, bad, good)
 Example: Wall Following
• How would you use feedback control to
implement a wall-following behavior in a robot?

• What sensors would you use, and would they

provide magnitude and direction of the error?

• What will this robot's behavior look like?

Feedback Control Terminology
• Plant P: process commanded by a Controller
• Process Variable PV: Value of some process
or system quantity of interest (e.g. temperature,
speed, force, …) as measured by a Sensor
• Set Point* SP: Desired value of that quantity
e(t) u(t)
SP - Controller Plant PV

Feedback Sensor
• Error signal e(t) = SP-PV: error in the process
variable at time t, computed via Feedback
• Control signal u(t): controller output (value of
switch, voltage, PWM, throttle, steer angle, …)
*Set point is sometimes called the “Reference”
 Bang-bang control

• Discrete on/off
• Furnace: goal
temp = 70
• when temp < 70
BANG! Heat;
• when temp >
70 BANG! Stop
the heat

Example source: Mathworks

 Bang-bang control
Desired speed Actual speed
Vd < V? *k motor
0/1
Encoder measurement

O(t) = k if v(t) < Vd

O(t) = 0 otherwise
Example: Home Heating System
• Plant P:
• Process Variable PV:
• Controller: Sensor:
• Set Point SP:
• Control signal:
e(t) u(t)
SP - Thermostat Boiler PV

Temperature Sensor
 Example: Home Heating System
• Plant P: Boiler with on-off switch (1 = all on ; 0 = all off)
• Process Variable PV: Current home temperature
• Controller: Thermostat Sensor: Thermometer
• Set Point SP: Thermostat setting (desired temp.)
• Control signal: Boiler on-off switch u(t) ∈ {0, 1}

e(t) u(t)
SP - Thermostat Boiler PV

Temperature Sensor

How could the function u(t) be implemented?

 Example: Home Heating System
• Plant P: Boiler with on-off switch (1 = all on ; 0 = all off)
• Process Variable PV: Current home temperature
• Controller: Thermostat Sensor: Thermometer
• Set Point SP: Thermostat setting (desired temp.)
• Control signal: Boiler on-off switch u(t) ∈ {0, 1}
e(t) u(t)
SP - Thermostat Boiler PV

Temperature Sensor

How could the function u(t) be implemented?

u(t) = 1 if e(t) > 0 [i.e., if SP > PV], 0 otherwise
 Motor Control: closed loop
• A way of getting a robot to achieve and
maintain a goal state by constantly comparing
current state with goal state.
• Use sensor for feedback

Desired speed Actual speed

test motor
computation

Encoder measurement
Motor Control: PID
• Control theory is the science that studies the
behavior of control systems
• CurrentState - DesiredState = Error
• Main types of simple linear controllers:
– P: proportional control
– PD: proportional derivative control
– PI: proportional integral control
– PID: proportional integral derivative control
Car Example
Proportional Control
• Suppose plant can be commanded by a
continuous, rather than discrete, signal
– Valve position to a pipeline or carburetor
– Throttle to an internal combustion engine
– PWM value to a DC motor
• What’s a natural thing to try?
– Proportional (P) Control: make the command signal
a scalar multiple of the error term: u(t) = KP  e(t)
e(t) u(t)
SP - Controller Plant PV

Sensor
 Example: Cruise Control (CC) System
• Plant P:
• Process Variable PV:
• Controller: Sensor:
• Set Point SP:
• Control signal:
e(t) u(t)
Vdesired - C-C System Engine Vactual

Speedometer
Example: Cruise Control (CC) System
• Plant P: Engine with throttle setting u ∈ [0..1]
• Process Variable PV: Current speed Vactual
• Controller: C-C system Sensor: Speedometer
• Set Point SP: Desired speed Vdesired
• Control signal: Continuous throttle value u ∈ [0..1]
e(t) u(t)
Vdesired - C-C System Engine Vactual

Speedometer

Define e(t) = Vdesired-Vactual, u(t) = KP  e(t), clipped to [0..1]

i.e. Throttle = KP  (Vdesired - Vactual)
Does this controller “settle” at the desired speed?
No; it exhibits error (E).
Proportional Control: Why E?

– Suppose e(t) = 0. Then u(t) = KP * e = 0 (Plant inactivated)

– Process Variable deviates from Set Point, activating plant
– But any real physical system has a delayed response
– Deviation, sustained over delay interval, yields error

Set Delay
Point
Error (E)
Process
Variable

PV < SP, u(t) > 0, Plant activates

PV = SP, u(t) = 0, Plant inactive
Time
Proportional Control Step Response
Notional plot and terminology:

Step
input Error (E)

Set Settle Time

Point
Process
Variable
Time

Is E constant over time? No; it depends on load.

When E = 0, plant turns off, so E grows, so plant turns on, etc etc.
This reaches an equilibrium with a constant E.
Proportional Control and Error
• Can combat E by increasing KP (“the P gain”)
• This gives a faster response and lower E!
• But increasing the gain too much leads to
overshoot and instability
Higher KP
Step
Set input
Point
Low KP

Process
Variable Time
Combating Overshoot: The D Term
• Note the derivative of error in responses below
• Subtract it from output to counteract overshoot
• Then u(t) = KP  e(t) + KD  d [e(t)] / dt
– KD the “derivative” or “damping” term in PD controller

Step Higher KP
Set input
Damped response, KD > 0
Point
Low KP
Small derivative
Process Large derivative
Variable Time

• … But still haven’t eliminated steady-state error!

Combating Steady-State Error: I Term
• Idea: apply correction based on integrated error
– If error persists, integrated term will grow in magnitude
– Sum proportional and integral term into control output
e(t) u(t)
SP -  Plant KP PV

  KI

Then u(t) = KP  e(t) + KI   e(t) (where the integral of

the error term is taken over some specified time interval)
This produces a proportional-integral (PI) controller

Incorporating the I term eliminates SSE by modulating

the plant input so that the time-averaged error is zero.
Putting it All Together: PID Control
• Incorporate P, I and D terms in controller output
– Combine as a weighted sum, using gains as weights

d/dt  KD

e(t) u(t)
SP -   Plant PV
KP
  KI

Then u(t) = KP  e(t) + KI   e(t) + KD  d [e(t)] / dt

This is a “proportional-integral-derivative” or PID controller
 Implementation Issues
• How do we approximate Ki *  err(t) dt to
implement an I controller?

• How do we approximate Kd * derr/dt to

implement a D controller?
PID Examples
Ziegler-Nichols Tuning Method
Exploration: set the plant under P control and start increasing
the Kp gain until loop oscillates
Note critical gain KC and oscillation period TC

KP KI KD
P 0.5KC
PI 0.45KC 1.2KP/TC
PID 0.5KC 2KP/TC KPTC/8
Z & N developed rule using Monte Carlo method
Rule useful in the absence of models
Critical Gain (Verge of instability)
Control summary
Control type Feedback Pro/Con

Bang- Simple
bang discrete yes /
Discrete

Open Simple/may
loop Control law no be
unrepeatable

Closed Continuous/
loop P, I, D yes Tune Gains
Velocity PID Controller

Here we are controlling wheel speed.

A rate controller leaves motor output unchanged if error is

0. No steady state error, so we can remove I.
What You Should Have Learnt!
• Early mechanical examples
• Feed-forward and Feedback control
• Terminology
• Basic controllers:
– Feed-Forward (FF) control
– Bang-Bang control
– Proportional (P) control
– The D term: Proportional-Derivative (PD) control
– The I term: Proportional-Integral (PI) control
– Proportional-Integral-Derivative (PID) control
• Gain selection
• Applications
• Car example
 Example: driving
• Steer a car in the center of a lane
 Example: driving
• Steer a car in the center of it lane

Observed error: distance off from center line

 Example: driving
• Steer a car in the center of it lane

Error is zero but how is the car pointed?

What will this do to the car?
P controller is happy on line independent of orientation!
What if respond ~ rate of change ?
• Steer a car in the center of it lane
 Example: driving
• Steer a car in the center of it lane

What is the observed rate of error?

Other error?
What if respond ~ rate of change ?
• Steer a car in the center of it lane

D controller is Happy on any parallel line!

 What if Road Sloped ?
• Steer a car in the center of its lane
 What if Road Sloped ?
• Steer a car in the center of it lane

Slower response uphill

Faster response downhill

Gravity contributes a steady-state error