robotics - australian national...

18
Robotics P@trik Haslum COMP3620/6320

Upload: others

Post on 28-Jun-2020

5 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Robotics - Australian National Universityusers.cecs.anu.edu.au/~patrik/publik/basicAI/robotics.pdf · Robotics Industrial Automation * Repetitive manipulation tasks (assembly, etc)

Robotics

P@trik Haslum

COMP3620/6320

Page 2: Robotics - Australian National Universityusers.cecs.anu.edu.au/~patrik/publik/basicAI/robotics.pdf · Robotics Industrial Automation * Repetitive manipulation tasks (assembly, etc)

Introduction

Robotics

Industrial Automation

* Repetitive manipulation tasks(assembly, etc).

* Well-known, controlledenvironment.

* High-power, high-precision,very expensive.

Mobile Robots

* Acting in environments thatare dangerous or difficult toaccess (Fire/rescue, mineclearing, Mars, deep oceans)

* Domestic (vaccuum cleaners,guides, toys)

* Unknown & dynamicenvironment.

* Uncertain sensors & impreciseactuators.

* Hard power/size/weight/costlimits.

Page 3: Robotics - Australian National Universityusers.cecs.anu.edu.au/~patrik/publik/basicAI/robotics.pdf · Robotics Industrial Automation * Repetitive manipulation tasks (assembly, etc)

Introduction

Page 4: Robotics - Australian National Universityusers.cecs.anu.edu.au/~patrik/publik/basicAI/robotics.pdf · Robotics Industrial Automation * Repetitive manipulation tasks (assembly, etc)

Introduction

Hardware

Sensors

* Range finders: Sonar, laser,contact sensors.

* Imaging: Cameras.

* Proprioceptive: shaftencoders, odometers /tachometers, inertial sensors(gyro, accelerometer),compass, force/torquesensors.

* External: GPS, beacons, fixedcameras.

Effectors

* Mobility: Wheels, tracks, legs,rotors / propellers, rudders.

* Manipulators: Arm joints,grippers (mechanical,vaccuum, magnetic), attachedtools.

* Speakers, light signals, etc.

Page 5: Robotics - Australian National Universityusers.cecs.anu.edu.au/~patrik/publik/basicAI/robotics.pdf · Robotics Industrial Automation * Repetitive manipulation tasks (assembly, etc)

Basic Problems & Techniques Localisation & Mapping

Two Basic Problems

Localisation & Mapping

* How does the world look, and where in it am I?

* Localisation: Determining the robots position/pose w.r.t. aknown environment.

* Mapping (SLAM): Constructing a map of the environment.

Path/Motion Planning

* Knowing the world and where I am in it, how do I get from hereto there?

* Path planning: Finding a path from A to B free of collisions withthe environment (obstacles).

* Motion planning: Finding a sequence of motions that take therobot from A to B without collision.

Page 6: Robotics - Australian National Universityusers.cecs.anu.edu.au/~patrik/publik/basicAI/robotics.pdf · Robotics Industrial Automation * Repetitive manipulation tasks (assembly, etc)

Basic Problems & Techniques Localisation & Mapping

Localisation

Filtering Approach

* P(xt) = P(zt |xt)∫

P(xt |xt−1, at−1)P(xt−1)

* Motion model (P(xt |xt−1, at−1)): deterministic prediction + noise.

* Sensor model (P(zt |xt)): likelihood of making observation z ifstate is x .

* Common assumption: Noise is Gaussian.

Page 7: Robotics - Australian National Universityusers.cecs.anu.edu.au/~patrik/publik/basicAI/robotics.pdf · Robotics Industrial Automation * Repetitive manipulation tasks (assembly, etc)

Basic Problems & Techniques Localisation & Mapping

Representation of the Estimate* Gaussian (“EKF”):- Compact and fast.- Unimodal, assumes linear models.

* Probability Grid (“Markov Loc.”):- Discretize configuration space, assign

probability to each cell.- Computationally expensive.

* Particle Filter (“Monte Carlo Loc.”):- Approximate distribution by a finite

sample of configurations.- Computationally expensive

(somewhat).

Page 8: Robotics - Australian National Universityusers.cecs.anu.edu.au/~patrik/publik/basicAI/robotics.pdf · Robotics Industrial Automation * Repetitive manipulation tasks (assembly, etc)

Basic Problems & Techniques Localisation & Mapping

Simultaneous Localisation and Mapping (SLAM)

* Filtering approach likelocalisation, but landmarkpositions part of the state.

* Problem: Dimensionality of statechanges dynamically as newlandmarks detected.

* Need to (reliably) reidentifylandmarks.

Demos by S. Thrun (http://robots.stanford.edu/)

Page 9: Robotics - Australian National Universityusers.cecs.anu.edu.au/~patrik/publik/basicAI/robotics.pdf · Robotics Industrial Automation * Repetitive manipulation tasks (assembly, etc)

Basic Problems & Techniques Path & Motion Planning

Two Basic Problems

Localisation & Mapping

* How does the world look, and where in it am I?

* Localisation: Determining the robots position/pose w.r.t. aknown environment.

* Mapping (SLAM): Constructing a map of the environment.

Path/Motion Planning

* Knowing the world and where I am in it, how do I get from hereto there?

* Path planning: Finding a path from A to B free of collisions withthe environment (obstacles).

* Motion planning: Finding a sequence of motions that take therobot from A to B without collision.

Page 10: Robotics - Australian National Universityusers.cecs.anu.edu.au/~patrik/publik/basicAI/robotics.pdf · Robotics Industrial Automation * Repetitive manipulation tasks (assembly, etc)

Basic Problems & Techniques Path & Motion Planning

Degree-of-Freedom (DOF)

* Degree-of-Freedom (DOF):Independent direction of robotmovement.

* Configuration (state/“pose”)specified by value for each DOF.

* Holonomic: # controllable DOF =# effective DOF.- Non-holonomic robot: Harder

planning/control problem.

Page 11: Robotics - Australian National Universityusers.cecs.anu.edu.au/~patrik/publik/basicAI/robotics.pdf · Robotics Industrial Automation * Repetitive manipulation tasks (assembly, etc)

Basic Problems & Techniques Path & Motion Planning

Work & Configuration Space

* Work space (W ): 3D world – Robot & environment have simplegeometry – Collision checking is easy.

* Configuration space (C): Space of robot states/poses(dimension = # DOF) – Robot is a point, obstacles havecomplex shapes.

* Free space: Configurations robot can reach/occupy.

* C −→ W easy, W −→ C hard & often ill-posed.

Page 12: Robotics - Australian National Universityusers.cecs.anu.edu.au/~patrik/publik/basicAI/robotics.pdf · Robotics Industrial Automation * Repetitive manipulation tasks (assembly, etc)

Basic Problems & Techniques Path & Motion Planning

Path Planning

* Search for a path in free space:Continuous space – need todiscretize.

* Cell decomposition:- Regular grid, subdivision, exact –

From incomplete to optimal.- Computationally expensive with

many dimensions.

* Skeletonisation methods.* Probailistic roadmap:- Scales to high-DOF problems.- Probabilistically complete, can yield

large “detours”.

Page 13: Robotics - Australian National Universityusers.cecs.anu.edu.au/~patrik/publik/basicAI/robotics.pdf · Robotics Industrial Automation * Repetitive manipulation tasks (assembly, etc)

Basic Problems & Techniques System Architechture

Robot System Architechtures

Traditional/Hierarchical

* “Sense – Plan – Act”: Build high-level world model,reason/plan in model, execute.

* Too slow, too “brittle” (sensing/acting failures).

Reactive

* “Sense – Act”: No explicit internal model.

* “Intelligence emerges from combination of simplebehaviours.”

* ...only sometimes, it doesn’t (inflexible, “dumb”).Hybrid

* “Sense – Act” & Monitor/Learn/(Re-)Plan/Adapt.

* 3-layer architectures (deliberative, reactive, control).

Page 14: Robotics - Australian National Universityusers.cecs.anu.edu.au/~patrik/publik/basicAI/robotics.pdf · Robotics Industrial Automation * Repetitive manipulation tasks (assembly, etc)

Applications & Success Stories

RHINO & Minerva: Robotic Museum Guides

* Demonstrated over ∼3 days in 1997 at Deutsches Museum inBonn, and 2 weeks in 1998 at the Smithsonian Museum inWashington.

* Navigation in an uncontrolled and crowded environment.

* Interaction with “untrained” users: Clearing the way, attractingattention, interpreting requests.

http://www.informatik.

uni-bonn.de/∼rhino/

http://www.cs.cmu.edu/∼minerva/

Page 15: Robotics - Australian National Universityusers.cecs.anu.edu.au/~patrik/publik/basicAI/robotics.pdf · Robotics Industrial Automation * Repetitive manipulation tasks (assembly, etc)

Applications & Success Stories

RoboCup* Robot soccer competition, held (almost) annually since 1997. In

2008, 70+ teams, in 4 leagues (+ other events/challenges).- Small (5/team, external camera/computers).- Midsize (4/team, all sensors/computers on-board).- Standard platform (Sony AIBO, Aldebaran NAO).- Humanoid (3/team, “kid” & “teen” sizes).

* Contributed to advancing s.o.t.a. in mobile robotics hardwareand programming – from “I see the ball” to real teamplay.

* “By the year 2050, a team of fully autonomous humanoid robots thatcan win against the human world soccer champion team.”

http://robocup.org/

Page 16: Robotics - Australian National Universityusers.cecs.anu.edu.au/~patrik/publik/basicAI/robotics.pdf · Robotics Industrial Automation * Repetitive manipulation tasks (assembly, etc)

Applications & Success Stories

Mars Exploration Rovers

* 2 Rovers, on Mars for over 5 years, havetravelled over 7.7 km / 15 km – still going!

* Daily “targets”, autonomous navigation &odometry, using stereo vision.

* Some “autonomous science” (rare eventscaptured in nav. cam images).

http://marsrovers.nasa.gov/home/

Page 17: Robotics - Australian National Universityusers.cecs.anu.edu.au/~patrik/publik/basicAI/robotics.pdf · Robotics Industrial Automation * Repetitive manipulation tasks (assembly, etc)

Applications & Success Stories

DARPA Grand Challenge 2005

* Challenge: Autonomously drivinga course of 212km,

* Route defined by coordinates(1 point / 72m).

* “Dirt road”, including some trickyspots.

* 5 vehicles finished the course (18didn’t), the fastest in < 7 hours(average of ∼30 kph).

http://www.darpa.mil/grandchallenge/

Page 18: Robotics - Australian National Universityusers.cecs.anu.edu.au/~patrik/publik/basicAI/robotics.pdf · Robotics Industrial Automation * Repetitive manipulation tasks (assembly, etc)

Applications & Success Stories

DARPA Urban Challenge 2007

* Challenge: Autonomously driving a 38“checkpoint” course on urban streets,given “network map”, with traffic, obeyingroad rules.

* 11 autonomous and 30 human-drivenvehicles simultaneously on the track – 6finished the course. * Note: The southern 6

waypoints in the Parking Lot (Zone 14) are Checkpoints 12 17

4-way Stop

Parking Lot*

Traffic Circle

2

8

1

117

6 4

5

9

10

3

1

2

3 4

6 7

89

10

11

13

12

5

14

WaypointLaneZoneStop Sign

Segment / Zone IDCheckpoint ID

Sample RNDF

1

1

v1.0

N

http://www.darpa.mil/grandchallenge/