L ECTURE 13: R EACTIVE CONTROLLERS B EHAVIOR -B ASED C ONTROL N - - PowerPoint PPT Presentation

16-311-Q INTRODUCTION TO ROBOTICS

LECTURE 13:

REACTIVE CONTROLLERS BEHAVIOR-BASED CONTROL NAVIGATION & OBSTACLE AVOIDANCE

INSTRUCTOR: GIANNI A. DI CARO

2

A CONTROLLER FOR PHOTOTAXIS BEHAVIORS?

We need to switch between two opposite behaviors What type of feedback controller should we use?

• The robot has two sensors, to the left and the

right side respectively

• Each sensor reports the amount of detected light
• If the amount of light, averaged from the two

sensors, is less than L, than the robot exhibits phototaxis: it heads toward the light

• If the amount of light, averaged from the two

sensors, is greater than L, than the robot exhibits anti-phototaxis: it escapes from the light

3

BANG-BANG CONTROLLERS: BINARY FEEDBACK

Wall following State 1: Too near - State 2: Too far Threshold: Binary output from a proximity sensor (wall y/n) Gain parameter: The ICR radius for moving to/away the wall Hysteresis: In absence of a distance measure, a time interval

• Two-states systems
• State transitions happen

depending on a threshold value

• The action at each state depends
• n a fixed gain parameter G
• Hysteresis can be helpful (in

general) to avoid over reacting State 1 State 2 >=Threshold < Threshold >= Threshold < Threshold Output(1, Gain) Output(2, Gain)

4

HYSTERESIS

Input Output

Out(1, Gain) Out(2, Gain)

Input ≥ Θ Θ ≦ Θ Δ ≥ Θ Θ Θ Δ Input < Θ ≥ Θ Θ

Input Output

Input ≦ Θ - Δ

Out(1, Gain) Out(2, Gain)

Input ≥ Θ Θ Θ - Δ Θ

No hysteresis With hysteresis State 1 State 2 >=Threshold < Threshold >= Threshold < Threshold Output(1, Gain) Output(2, Gain)

Bang–bang controls with hysteresis provide optimal controls in some cases, although they are

• ften implemented just because of

their simplicity or when binary behaviors are required

Braitenberg vehicle basics Implementation as simple electronic circuits

• Sensors directly connect to motors
• Can increase or decrease motor activity
• Behavior depends on
• geometrical arrangement of sensors
• geometrical arrangement of motors
• cross connection of sensors
• Emergent behaviors
• Competing stimulation / inhibition creates complex actions

5

B R A I T E N B E R G V E H I C L E S

Basic vehicle ∼ One single neuron / motor that fires proportionally to the external input (from the sensor) Activation can be + or - Two or more “neurons” can be put together …

6

B A S I C T Y P E S

Basic three types of Braitenberg vehicles

• Sensors stimulate same-side motors
• Sensors stimulate opposite motor
• Sensors go to both (all) motors, some

stimulating, some inhibiting

7

O B S TA C L E AV O I D A N C E

• Sensors generate proximity value
• The closer to an object, the higher the sensor output
• Obstacle left
• Obstacle right
• Obstacle center

8

L I G H T S E E K I N G

• Sensors generate light value
• The more light, the higher the sensor output
• Obstacle left
• Obstacle right
• Obstacle center

9

C O M P L E X B E H AV I O R S

Braitenberg: Light seeking + obstacle avoidance Depending on weights, different behaviors

• Following a path
• Docking to a charger station
• Coordinated moments of a swarm of robots
• ... and many more

10

F R O M A N O B S E R V E R P O I N T O F V I E W …

Out of simple rules (and possibly imprecisions in actuations that naturally break symmetries) a number of different behaviors can be displayed that, to an external

• bserver, could look like expressions of some personality, emotion, social interaction …

11

F O R M O R E I N F O …

https://www.youtube.com/watch?v=A-fxij3zM7g A nice video talking in general about Braitenberg’s book and his vehicles + nice small robots programmed to act as the vehicles and show a complex world …

12

REACTIVE PARADIGM FOR ROBOT CONTROL: DON’T THINK, REACT!

Starting from the mid 1980s, a number of different views (mostly bio-inspired) and approaches were developed and employed in robotics (and in AI, that started moving from symbolic to sub-symbolic / neural models) Deliberative paradigm Internal models Look ahead

Functional pipeline

Top-down problem solving Requires a closed world Reactive paradigm Model free React to inputs

Concurrent modules

Bottom-up problem solving Works in an

• pen world

13

Sense

Model

Plan Act Environment

ARCHITECTURE BASED ON REACTIVE PARADIGM

Environment

Sense Act

Rule 3 Rule n Rule 1 Rule 2

Sense-Act Transfer rules Behaviors

Vertical decomposition vs. Horizontal decomposition

Ethological view (Behavior): Direct mapping of sensory inputs to a pattern of motor actions that are then used to achieve a task Mathematical view (Function): A transfer function, transforming sensory inputs into actuator commands

Concurrent mode vs. Sequential mode

14

EXAMPLES OF SENSE-ACT RULES

Environment

Sense

Identify Objects Monitor Changes Explore Wander Avoid Objects

Act

Environment

Sense

Build Map Find Path Track Person Follow Wall Avoid objects

Act

15

Good reference books, plenty of ideas and discussions on the Reactive / Behavioral paradigm:

QUESTIONS TO CLARIFY

• 1. Where the Reactive paradigm finds its roots?
• 2. What is the exact nature/characteristic of the SENSE-ACT rules?
• 3. How the ACTion output from the different rules is arbitrated to result into

a single, coherent Action command to the Effectors?

• R. Arkin, Behavior-Based Robotics, MIT Press, 1998
• R. Murphy, An Introduction to AI Robotics, MIT Press, 2000
• M. Mataric, The Robotics Primer, MIT Press, 2007
• J. Jones, A Practical Guide to Behavior-based Robotics, McGraw-Hill, 2004

16

THE BIOLOGICAL ROOTS

Many “simple” animals exhibit individual and collective intelligent behavior yet have virtually no

• brain. Therefore, they must be doing something

to manage world’s representation complexity!

Animals live in an open world, and roboticists would like to overcome the closed world assumption

Historical track: Cybernetics (back to the 40’s, N. Wiener, G. Walter’s Turtoises),

• V. Braintenberg’s (conceptual) vehicles based on direct SENSE-ACT
• 1. Where the Reactive paradigm finds its roots?

Dissatisfaction with the limitations of the Deliberative approach

17

REACTIVE RULES AS BEHAVIORS

• 2. What is the exact nature/characteristic of the SENSE-ACT rules?

A fundamental building block of natural intelligence is a behavior: a mapping of sensory inputs to a pattern of motor actions, which then are used to achieve a task Ethology: study of animal behaviors

Sensor Inputs Pattern of Motor Action

Behavior

Releaser

Sensor input: Water source detected Releaser: Giraffe is thirsty Releaser: No predators Action pattern:

Move head checking for predators Put legs in right position, Lower the neck Adjust legs position Drink rapidly Neck up and check surroundings

Task: Drinking

18

CATEGORIZATION OF (ANIMAL) BEHAVIORS

Reflexive behaviors: Stimulus-response (S-R), “hardwired” behaviors. The stimulus is directly connected to the motor action to produce the fastest response time. No cognition: if you sense it, you do it!

Arctic tern feeding: When hungry, babies peck at the closest red blob

Source: R. Murphy, AI Robotics, MIT Press, 2000

• R. Arkin, Behavior-Based Robotics, MIT Press, 1998

Increase of complexity

In ethology, a reactive behavior means a learned behavior or a skill In robotics, the word reactive (mostly) connotes a reflexive behavior.

Conscious behaviors: Deliberative, requiring conscious thought, possibly combining previously developed behaviors

Assembly a robot?

Reactive behaviors: Learned, and then consolidated to where they can be executed without conscious thought, but can be changed by conscious thought.

Hunting “Muscle memory”

19

CATEGORIES OF REFLEXIVE (ANIMAL) BEHAVIORS

Taxes

Response: to move to a particular orientation marked by a stimulus intensity Phototaxis Chemotaxis

Reflexes Stimulus t1 ——— t2 Response t1 ——— t2

Value Axis 50 100 Category Axis AprilMayJune July

Stimulus intensity Response

Short/Instantaneous

Fixed-action patterns Stimulus t1 — t2 Response t1 ———— t3 t2

Keep fleeing after predator detection Persistent

20

FROM ANIMAL TO ROBOT BEHAVIORS

• A. Ethology / Animal behaviors have been, are, and will be a major source
• f inspiration to design effective robotic behaviors!
• B. As a matter of fact, in terms of taxonomy, design principles, architectural
• rganization, the reactive paradigm in robotics has mirrored very much

the results coming from the ethology field

• C. What we have discussed so far has been translated /adapted into a

number of different views of the reactive paradigm: Subsumption, Motor schema, Potential fields, Behavior-based control, … Why all this discussion about animal behaviors when we are (mainly) interested to robot behaviors?

21

A PRACTICAL EXAMPLE: OBJECT COLLECTION

The robot:

• Frontal IR emitters / detectors
• Light sensors
• Frontal bumper
• Two standard wheels

Does it look like as a simple task to program/control?

• The robot searches for type A objects
• When an A object is found, it has to be

brought (by pushing) at a storing location identified by a bright light

• The robot has to collect as many A
• bjects as possible
• Other objects cannot be pushed
• The environment can feature walls

Robot

Storage

A A A A A A A A

Task specification:

22

BEHAVIOR-BASED ARCHITECTURE FOR OBJECT COLLECTION

Environment

Sense

Escape No dark-push Anti-moth Go to objects Home

Motor control

Cruise

IR

detectors

Light

sensor

Bumper

force

Arbiter Priority

From: J. Jones, A Practical Guide to Behavior-based Robotics, McGraw-Hill, 2004

Bottom-up design: try to devise a set of simple behaviors that, when acting together, produce the overall desired activity

↓

Emergent (global) behavior

A A A A A A A A

23

BEHAVIOR ARBITRATION

• 3. How the ACT output from the different rules is arbitrated to result into a

single, coherent Action command to the Effectors?

Environment

Sense Act

Behavior 3 Behavior n Behavior 1 Behavior 2

A r b i t r a t i

• n

Conflict resolution module

It might be needed also to access / adjust sensory systems

24

BEHAVIOR ARBITRATION STRATEGIES

Averaging / Composition B1 ⊕ B2 Voting {R1, R2, R3}: X {R4}: Y

⇒ X

Least Commitment {R1}: DON’T X Fixed priority B1(t) ≻ B2(t), ∀t Alternate B2([t1,t2]), B1([t2,t3]) Variable priority B1(t1) ≻ B2(t1), B2(t2) ≻ B1(t2), Subsumption Suppression: BNew ≻ BOld Inhibition: BNew ⋀ BOld ⇒ ∅

25

BEHAVIORS FOR MOVING TO A POINT IN PRESENCE OF OBSTACLES

Bug 1 and Bug 2 Insect-inspired Algorithms (1987)

• Local decisions, map-free!
• Complete algorithms: find a

path if it exists, otherwise returns with a failure.

• No guarantees in terms of

finding a good (short) path!

• “Nasty” obstacle structure and

placement or “wrong” left/ right turning choices can make things going arbitrarily bad.

26

BUG 1: ~EXHAUSTIVE SEARCH

cannot

27

BUG 2: GREEDY SEARCH

28

B U G 2 : I S S U E S

29

B U G 1 O R B U G 2 ?

30

(STRONG) ASSUMPTIONS

Localization: The robot must know where it is in the world, and must be able to maintain a (robust) estimate of this while it moves Mapping: The location of the goal is known within the robot’s representation

• f the world.

Mapping/Localization: The robot must be able to mark a location in its representation of the world and be able to determine that it has returned to it. Sensing: Obstacles need to be sensed with precision, in order to circumnavigate them. Actuation/Control: The robot needs to be able to smoothly change and adapt its trajectory … velocities?

31

G E N E R A L O B S TA C L E AV O I D A N C E

Dynamic Window Approach (DWA, 1987) ICR

• D. Fox, W. Burgard, S. Thrun, The Dynamic Window Approach to Collision Avoidance, IEEE Robotics & Automation Magazine

4(1):23 - 33 · April 1997

32

D WA : F E A S I B L E F R E E S PA C E V E L O C I T I E S

Vs = Velocities admissible by robot kinematic constraints

33

D WA : A D M I S S I B L E V E L O C I T I E S I N T I M E W I N D O W T

Time interval

34

D WA : A D M I S S I B L E V E L O C I T I E S G I V E N O B S TA C L E S

Admissible velocities (allow the robot to stop without colliding with obstacle)

Decelerations for breakage

• nce reaching the obstacle

Distance to the closest

• bstacle on the circular

trajectory defined by (v,𝜕)

35

O B J E C T I V E F U N C T I O N / B E H AV I O R

Clearance from obstacles Target reaching Velocity (fast, smooth)

36

O B J E C T I V E F U N C T I O N / B E H AV I O R

Admissible velocities given accelerations dvb/dt = 50 cm/s2, d𝜕/dt = 60 deg/s2 Non admissible velocities are dark shaded areas. E.g., all velocities in area right wall II would cause a sharp turn to the right and thus cause the robot to collide with right wall.

Admissible velocities in the time window are in the white area