Mobile Robot Controllers

• Brain of the robot

• Three architectures:
• Sense-Plan-Act (old school)
Mobile Robot Control Techniques • Subsumption (Rodney Brooks)
• Three-Layer (Modern)

Processor
Sensors

Dr. Rohan Munasinghe

BSc, MSc, PhD, MIEEE
Department of Electronic and Telecommunication Engineering
Faculty of Engineering
University of Moratuwa 10400
Actuators
Sensors

SPA: Sense - Plan – Act Robots Have Many Goals

• Consists of 3 linear, repeated steps:
– Sense your environment
I need to go
– Plan what should be done by building a world to the office
model (where am i? whats near me? where is the
Sense I need
destination) through sensor fusion (combining a more energy
number of sensors), and considering goals and
constraints Plan
Objectives
Red traffic
• short term: Avoid eminent collisions light turned on I am about to
• long term : Reach the destination
Act I just want fall over
to be loved
– Action: Execute the plan through the actuators

• SPA was the predominant robot control

mechanism till 1985 A goal’s priority will change based on context or present situation
 Slicing the SPA SPA
Sensors • Simple, straight forward, easy to implement
• All goals are known at
each stage, and affect • Appropriate for simple tasks
the computation Perception
• Monolithic design: Sequential process
Modeling
• At each step, we have to do sensor fusion, world
Planning modeling, and planning for all goals, before executing an
action
Task Execution
• Applicable for simple mobile robot applications in
Motor Control structured or moderately unstructured environments that
can be coped up with few sensors and actuators

Actuators

Subsumption Architecture Subsumption Architecture

• Introduced by Rodney Brooks (MIT) in 1986
reason about object behavior
• Consists of layered behaviors, from simple to complex,
plan changes to the world with simple interfaces between them

monitor changes • Layers can “override” each other

Sensors Actuators
build maps • Each layer has a control program that is capable of
working at the speed of environmental change
explore – Speeds up robot response

wander • Each layer now can do the appropriate model building,

sensor fusion, etc.
avoid objects
 Subsumption Example Layer Subsumption
Behaviors: Touch, Turn, Wander
Touch has the highest priority, followed by Turn and then Wander. This • Each layer has one function, conceptually
ordering makes sense because while the robot’s wandering about, if it hits
something, the obstacle avoidance should take precedence over the wandering. • Lower layers tend to be more “reactive”
• closed loop controls
• inputs tightly coupled to outputs

• Higher layers are more “deliberative”

• Performs high-level sensor fusion and world modeling
• Predicts future scenarios, and plan a course of motion

• Layers can take the inputs and/or outputs of other

layers

Subsumption Advantages Three-Layer Architectures

• Provides a procedure to incrementally build and test • TLA was developed in response to the subsumption
a complex mobile robot control system architecture
• Layers can be inserted, removed, or revised
• TLA design seems to implicitly:
• Supports parallel computation in a straightforward,
intuitive way • Agree that different subsystems (layers) are needed to
• Appropriate for sophisticated robots that inhabit complex react to events on different time scales
environments
• Agree with loose asynchronous interfaces
• Avoids centralized control; relies on self-centered and
• Disagree with the “infinite regression” of layers
autonomous modules
• Disagree with the subsumption mechanism, i.e. overriding
of inputs/outputs
 The Three-Layer Architecture Feedback Control
• Example: Home Heating system
Consists of 3 layers • Thermostat measures room temperature, and compares it with the
desired temperature, and calculate the error
• Reactive layer (Controller)
• Use the error signal to drive the heater in order to reduce the error
• Sensor-based, short-time scale actions towards zero
• For collision avoidance etc.. • If error is +ve turn on the heater
• If error is –ve turn off the heater
• Heater is a binary (ON/OFF) heat source
• “Glue” Layer (Sequencer)
• There is a delay in air heating up and cooling down, which causes
• Has a memory of the past, and selects primitive
air temperature to oscillate around desired temperature
actions for controller

• Planning Layer (Deliberator)

• Plans for the future
• Time-consuming operations

14

Proportional Control of Motors Proportional Control of Motors

Generates a stronger control signal when the system is farther away from the
Overshoot:
goal state — control signal is proportional to the amount of error
Response moves beyond set-point, and stops and turns back.

At t=0, suppose that the setpoint =100 and actual position=0. Then
e(0)=100, so the motor turns at 100% power, driving the wheel toward the Oscillations:

desired position. As it moves, the error becomes progressively smaller. After an overshoot, error becomes –ve, thus, proportional controller drives

When it’s halfway at position=50, the error is only 50, and the motor goes at the motor in the opposite direction. In the subsequent motion, motor might

50% power. When it arrives at the intended position of 100, the error is zero, “undershoot” as well. This overshoot and undershoot phenomena repeats

and the motor turns off momentarily. for while, and gradually, the oscillation is expected to die out.

Proportional Gain (ratio between error and power): Instead of a one-to- Controller Design:
one ratio between error and motor %power, modify the controller so it Design to minimize both overshoot and oscillation, but let little overshoot to
multiplies the error value by 5. The wheel should reach the set-point faster, improve system response to reach the set-point as quickly as possible.
and it should resist being turned away from it much more aggressively. 15 16
 Proportional Gain Kp Proportional Gain Kp
Kp =10: Full-power is delivered Kp =30
to motor as long as error>10.
When position falls below 10, power Kp =10 Causes predominant oscillation Kp =30
command began to fall off. System
overshot the zero point, and shows
little oscillation Kp =50
Steady State Error (Offset): Oscillation behavior has taken over.
System does not stabilize at the
set-point, but at 1. This System is too sensitive (responsive). A
generates a power command of slight error causes huge power delivered
10%, which is too small to Kp =50
activate the motor through the motor. Subsequent slight
Kp =20 errors in +ve and –ve directions causes
Kp =20: should ameliorate the swing sustainable swing around the set
offset problem, since the same point . While the position error is small
static error will result in a higher
power command on the graph, the power command
Offset has been eliminated, but swings are quite evident.
at cost of oscillations before 17 18
settling

Proportional-Derivative Control Proportional-Derivative Control

• Observation/Problem: Trade-off between offset and oscillation
Kp=4, KD=1:
Cranking up Kp alone could not produce higher performance. Higher Kp Kp=4, KD=1
Overshoot is minimized, and there is
drives the wheel to the set point faster, but results in overshoots and
no any significant oscillatory behavior.
oscillates at higher gains

Kp=10, KD=5:
• Solution: Use derivative control
Observe this phenomena: At larger errors, velocity is small, and at KD is too large. Controller puts on the
Kp=10, KD=5
smaller errors velocity is large (as motor swings around set point) brakes too aggressively and the
system stops too early before reaching
• What if we reduce motor power by a quantity ∝ speed.
the set-point. When the velocity hits
• No effect at larger errors
zero, the proportional gain kicks in
• Reduce power near set point (reduce oscillations)
again and the system corrects
This looks like what we need to improve response while reducing
19 20
overshoot