robosoccer team mi20 presents … supervisors albert schoute mannes poel current team members paul...

Post on 12-Jan-2016

215 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

TRANSCRIPT

Robosoccer Team MI20 presents …

Supervisors• Albert Schoute• Mannes Poel

Current team members• Paul de Groot• Roelof Hiddema

Mobile Intelligence Twente

Robot soccer as a scientific “playing field”

Interdisciplinary• Hardware & Software• Sensing & Control• Image processing • Motion planning• Multi-agent collaboration• Communication• Artificial intelligence

International• Championships

(FIRA, Robocup)• Congresses

Mission Impossible ?

International leagues

Robocup• Humanoid• Small size• Middle size• Four-Legged• Rescue• Junior• Simulation• @Home

FIRA• HuroSot• KheperaSot• MiroSot UT team• NaroSot• QuadroSot• RoboSot• SimuroSot

Robocup Humanoid 2 vs. 2 (Osaka 2005)

FIRA Humanoid (Vienna 2003)

Robocup Middle League

Robocup Small League

FIRA Mirosot (11 vs. 11)

• Games between teams of 5, 7 or 11 robots• Camera’s above the field observe the playing• Computers control the robots wirelessly

FIRA Mirosot competitie

MiroSot robots

• Maximal dimensions:7.5 x 7.5 x 7.5 cm

• Two-wheeled differential drive robots

• Board-computer controls wheel velocities

Impression of EC 2005 @ UT

Twente’s robosoccer team

• Started in 2002:

Missing Impossible

Mission Impossible

Mobile Intelligence

Generations of students

1st teamLjubljana 2003

4th teamVienna 2006

Generations of robots

Home base

Computer control

Localization

• Robots have color patches on top

• Design is free, except for obligatory team color

• Design choice:identical or different patterns per robot?

• Identical makes recognition simpler, but robots must be tracked

Vision

Camera image

Color segmentation

Color separation

Region clustering

Camera calibration

• Lens distortion

Image correction

• Remap feature points only

Correction of projective mapping

• Automatic field calibration by 4 known markers

Correct for parallax

Tracking

State estimation

y

x

v

v

y

x

y

x

y

x

State

) θx

y

(x,y)

Result on the screen

Motion Control

• Robots have local PID velocity controllers• Motion commands wheel speeds (vr, vl) cq. lin.

& ang. velocities (v, )• Kinematic robot model

• Higher speeds: account for dynamics!

10

0sin

0cos

y

x

Motion Planning

Driving fast to play the ballwhile avoiding obstacles …

Strategy ?

The team’s magic

System design ?

PlayerAgentPlayerAgentPlayerAgentPlayerAgent

prediction &derivation

Camera

Coachagent

UserInterface

PlayerAgent

Robot

TimageTsettings

Tvision_settings & Tcalibration

Tgame_status & Tscore_board

Tscore_board & Tgame_status

Tworld_data

Tworld_data

Tworld_data

Tcoach_data

Tsensor

Tsensor & Todometrics

Twheel

Motors

State

Tvoltage

Tsensor & Todometrics

Todometrics

User

RFcommunication

Twheel

Tsensor & Todometrics

Vision Measurement

Tsnap_shot

Simulator DLL

Tsnap_shot

Twheel

Tsensor & Todometrics &

Timage

Tgame_status & Tscore_board

Tplayer_data

Tscore_board & Tgame_status

The team’s pain

(Re)designing for the winning team

Initial MI20 multi-agent system architecture:

1st team motion controller

Solve the parking problem:

move to “pose”(x, y, )

… while avoiding obstacles

Vector Field Histogram Corresponding Histogram

Local method:

Trying out in de simulator

Shoot and score!

Shoot and miss!

Improvements

Real Prediction

1

2 2

1

1

2

2

1

Avoid tracking errors by collision analysis

Collision prediction

last1

pred1

pred2

γ

last2

Collision state correction

corr1

corr2last2

pred2

last1

pred1

Collision response model

B

VB

P

n

A

VA

Collision response (cont.)

B

VBωBA

VAωA

P

Improving strategy

Choosing optimal offensive / defensive positions

Improved system structure

• Complete software revision

• Reduced thread concurrency

• Simplified interprocess communication

• Current O.S.Linux Fedora Core 4

Coming soon …

TU Vienna Parade

Questions?

top related