Design Review

79
Webots Competition Tommaso Buvoli, Patrick Cromer, Sean Egan, Nicholis Marsiglia, Kelsey Metzler, Srinivas P

description

 

Transcript of Design Review

Page 1: Design Review

Webots Competition

Tommaso Buvoli, Patrick Cromer, Sean Egan,

Nicholis Marsiglia, Kelsey Metzler, Srinivas P

Page 2: Design Review

Strategy - Sean

Stationary goalie until opposing players come within a

certain range

Then, side-to-side motion

Outside of a certain range of the opposing goal, line up

the shot and kick it

Opposing goalie avoiding technique

Two players other than goalie are offense/defense

Offense: Closer player to ball, kicks towards goal

Defense: Knock over other players or block other players

Page 3: Design Review

Locomotion - Patrick

Implement StandUpFromBack.motion

(Based off of StandUpFromFront.motion)‏

Implement odometer model for existing motions

Improve walking motion

Page 4: Design Review

Improved Walking

Make forward motion quicker and/or more stable

Use a central pattern generator (CPG) for walking

CPG: "neural circuits that generate periodic motor

commands for rhythmic movements such as locomotion." [Kuo,

Arthur D. "The relative roles of feedforward and feedback in the control of rhythmic movements." Motor Control (2002): 6, 129-145.]

Page 5: Design Review

We are using Grid based approach with Kalman filtering.

Kalman filter localization tracks the robot from an initially

known position and is inherently both precise and

efficient.

Picture from: www.cyberbotics.com

Localization - Srinivas

Page 6: Design Review

Vision – Kelsey, Nick, Tommaso

We will work on researching and implementing

appropriate vision filters for the task

Edge Filter

Color Filter

Shape Filter

Page 7: Design Review

Edge Filter

Used to detect the GOAL

The goal is comprised of straight, vertical and horizontal lines

The net of the goal also has straight lines

Scale Invariant Feature Transform (SIFT)

Possible Issues

The field also has painted straight lines

To Remedy: look for VERTICAL and HORIZONTAL edges

Page 8: Design Review

Color Filter

Used to detect GRASS

Grass is the only green thing in the game

Another way to avoid thinking the field is the goal

Color oriented moving object detection

Possible Issues

What if another team makes a green player?

The Remedy: identify players based on shape

Page 9: Design Review

Shape Filter

Used to detect PLAYERS

Players have a distinctive shape

Helps avoid color issues

It is necessary to detect both your players and other players

Speeded Up Robust Features (SURF)

Possible Issues

May be difficult to program in the shape of the player

Other random objects may be detected as player-shaped

Page 10: Design Review

Shape Filter

Used to detect BALL

Ball is spherical, should be a circle from any angle

Easier than color pattern detection for ball

Necessary to know where ball is to kick it

Speeded Up Robust Features (SURF)

Possible Issues

If ball is moving very quickly, it may be more of a blurred oval

than a circle

This can likely be accounted for

Page 11: Design Review

Camera Calibration

Page 12: Design Review

Affine Transforms

Page 13: Design Review

Feature Recognition

Page 14: Design Review

Robostadium Project:Presentation 1

Group 2

Nisheeth Bhat

Mason Hauck

Ken Hua

Chris Karman

Niket Sheth

Vishal Verma

Page 15: Design Review

Introduction

• Locomotion• How can we move around the field more effectively?

• Perception• Where's the ball? Where's the goal? Where are the other

players?

• Communication• How do we coordinate multiple robots effectively?

• Localization• Where is the robot on the field?

• Planning• What should each robot be doing based on input and

strategy?

Page 16: Design Review

Locomotion

• List of Available Motions:• Backwards• Forwards• Forwards 50• Hand Wave• Shoot• Side Step Left• Side Step Right• Stand Up From Front• Turn Left 180• Turn Left 40• Turn Left 60• Turn Right 40• Turn Right 60

• Additional Motions:• Rotate Around Ball• Diagonal Step• Roll or Stand Up From

Back• Tackle?• Fall Left (Goalie)• Fall Right (Goalie)• Fall Forward (Goalie)• Jump (Goalie)• Catch Ball (Goalie)• Expand Shoot so that it

can kick diagonally and backwards for passing the ball?

Page 17: Design Review

Perception

• Sensors• Tilting Camera

• Ultrasound sensor

• 2D gyroscope

• Bump sensors

Page 18: Design Review

• Color sensing using thresholding for the required color

• Detect the forms of various objects in the field we will use edge detection• Edge detection will be implemented using the

appropriate filters/kernels like Sobel

• Object recognition will be done by Hough Transform

• Detect other players

• Detect Ball (orange color)

• Detect Goal

Tilting Camera

Page 19: Design Review

• Measure the actual distance to the sensed object• Distance to ball

• Distance to other player

Ultrasound Sensors

Page 20: Design Review

Determine changes in orientation of the object

Detect when the robot falls down

2D Gyroscope

Page 21: Design Review

Detect collision between two objects

Detect a successful kick

Bump Sensors

Page 22: Design Review

Communication

• Boolean Values• See ball

• See goal

• Can shoot

• Can pass

• Ready to receive pass

• Fallen down

• Integer Values• Player ID

• Player ID of passing target

• Number of teammates in view

• Number of opponents in view

Page 23: Design Review

Communication

• Float Values• Distance to ball

• Distance to goal

• Player heading

• Array Values• Player location

• Ball location

• Teammate locations

• Opponent locations

Page 24: Design Review

Localization

• Divide into 2 levels:• Macro Level:

• Decide current Macro-block

• Use global techniques such as landmarks

• Micro-Level:

• Decide on location within said Macro-block

• Use local techniques such as odometry

Page 25: Design Review

Planning1. Check for ball

2. Check for team w/ ball

3. Find ball

4. Closest player to ball?

5. Go to ball until at ball

6. Find goal and distance

7. Check for obstacles

8. Kick

9. Check teammate with clear path

10. Relocate ball

11. Pass

12. Position to clear path to goal

13. Wait for other player to pass

14. Check for score

15. Stop all players

Page 26: Design Review

Questions?

Page 27: Design Review

March 11, 2010 Group#3

RobotStadium ProjectGroup #3:

Michelle BourgeoisNicholas Farrow

John MartinJoe McCabeMichael PackZach Pranger

Page 28: Design Review

Overview

MotionSensingLocalizationCommunication / PlanningStrategies

Page 29: Design Review

March 11, 2010 Group#3

Motion: Getting Up From Back

1. Goal is to flip over to front2. Use accelerometer to check if flip was successful3. Use given get up from front motion

Page 30: Design Review
Page 31: Design Review

Motion: Fix Standing Up From FrontCurrent motion file not workingModified .motion file

Page 32: Design Review

Motion: Running

Goals for running:Must not compromise stabilityMove faster than normal walk

Options for running:Use motion editor to generate motion fileOptimize the walking

Maybe remove unnecessary transitional motionsStraight legs vs bending knees

Page 33: Design Review

March 11, 2010 Group#3

Motion: Walking In an Arc

Currently 1. Robot looks for the ball 2. Turns to walk straight 3. If the ball moves, it stops to reassess

How do we plan to do this? Continue walking while reassessing Left foot steps further than right OR Right foot steps further than left If ball is in front(±5 degrees), walk straightElse, launch turn and walk

Page 34: Design Review

Overview

MotionSensingLocalizationCommunication / PlanningStrategies

Page 35: Design Review

March 11, 2010 Group#3

Sensing: Camera Vision

How to detect the goal and ball:1. Run through the RGB color channels2. Look for color changes with a threshold variable3. Mark changes with a 1 in a new matrix4. Build a new matrix with an outline of the goal5. Align player to goal and shoot

Page 36: Design Review
Page 37: Design Review

March 11, 2010 Group#3

Sensing: Range FinderFour rays (upper left, lower left, upper right, and lower right)Range 0.0 to 0.7 metersUse to detect obstacles (other players) Incorporate in path planning

Page 38: Design Review

March 11, 2010 Group#3

Sensing: Foot Bumpers

detect successful contact with ball (kick)0 or 1 Boolean value

when a kick command is issuedif bumper value = 0

ball should be near feetelse bumper value = 1

ball should be out aheadadjust head pitch to likely ball location

Page 39: Design Review

Overview

MotionSensingLocalizationCommunication / PlanningStrategies

Page 40: Design Review

March 11, 2010 Group#3

Localization

Page 41: Design Review

March 11, 2010 Group#3

Localization

Page 42: Design Review

March 11, 2010 Group#3

Localization

Page 43: Design Review

March 11, 2010 Group#3

Localization

Page 44: Design Review

March 11, 2010 Group#3

Localization

Page 45: Design Review

March 11, 2010 Group#3

Localization

Page 46: Design Review

March 11, 2010 Group#3

Localization

Page 47: Design Review

March 11, 2010 Group#3

Localization

Page 48: Design Review

March 11, 2010 Group#3

Localization

Page 49: Design Review

March 11, 2010 Group#3

Localization

Page 50: Design Review

March 11, 2010 Group#3

Localization

Page 51: Design Review

March 11, 2010 Group#3

Localization

Page 52: Design Review

March 11, 2010 Group#3

Localization

Page 53: Design Review

March 11, 2010 Group#3

Localization

Page 54: Design Review

March 11, 2010 Group#3

Localization

Page 55: Design Review

March 11, 2010 Group#3

Localization

Page 56: Design Review

March 11, 2010 Group#3

Localization

Page 57: Design Review

March 11, 2010 Group#3

Localization

Page 58: Design Review

Overview

MotionSensingLocalizationCommunication / PlanningStrategies

Page 59: Design Review

March 11, 2010 Group#3

Communication / Planning

Record player & ball locationstriangulate ball position between playersconfirm player positions (both teams)

Planning

send a player to shoot the ball possibly send a player to defend the goal path planning around other playersfeedback information from sonar sensors

Page 60: Design Review

Overview

MotionSensingLocalizationCommunication / PlanningStrategies

Page 61: Design Review

March 11, 2010 Group#3

StrategiesStrategy 1

Closest to the ball

Strategy 21 player shoots, 1 player defends

Strategy 3Player with the best alignment shoots

Goalkeeper:

Position itself aligned with the ball If ball in range, dive

Page 62: Design Review

Best Aligned Player

Page 63: Design Review

Questions?

Page 64: Design Review

Robot Stadium

Page 65: Design Review

Sensors

• Camera

– Boundary Detection

– Player Detection

– Distance to nearest opponent– Distance to nearest opponent

– Distance to nearest teammate

• Ultrasound

– Only useful within 0.7 m

– Useful in some instances, not others

• Players would be too late, ball and goal might help

Page 66: Design Review

Sensors II

• Gyroscope

– Help with position by measuring turning rate

– May be too much noise to be reliablereliable

• Accelerometer

– Can tell whether the robot fell on its front, side, or back

– Can tell when the robot is standing again

Page 67: Design Review

Sensors III

• Foot Sensors

– Can tell if the robot is standing on its feet

– An alternate way to see if the robot has gotten up again.robot has gotten up again.

• Foot Bumper

– Good way to tell the robot kicked the ball

– Also a way to find out the robot tripped

Page 68: Design Review

Locomotion

� Recovering From a Fall

� Turning While Running

� Kicking the BallKicking the Ball

� Defensive Falling

Page 69: Design Review

Recovery

� Getting up when on your stomach.

� Getting up when on your back.

� Getting up when on your side.Getting up when on your side.

� Rolling over.

Page 70: Design Review

Turning while running

� More versatile

� Faster reaction time

� Can get to the ball soonerCan get to the ball sooner

Page 71: Design Review

Kicking the Ball

� Kicking the ball instead of running into it

� Longer range, more accurate

� Purposeful offense instead of accidentalPurposeful offense instead of accidental

Page 72: Design Review

Defensive Falling

� Falling over could be used on purpose to:

− Block the goal

− Block the ball

Create an obstacle for other players− Create an obstacle for other players

Page 73: Design Review

Particle Filter

• Using mostly cameras and some ultrasound.

– Base robots’ positions on size in camera

• Require experimentation to get usable scaling

• Be necessary to determine the FOV of the camera• Be necessary to determine the FOV of the camera

– Ultrasound near goal and near other robots

• This sensor would have to have high priority updating

Page 74: Design Review

Particle Filter Sample

Page 75: Design Review

Planning

� Direct Path (A → B)� A*

� Offensive/Avoid

PathPath� A* biased w/ Voronoi

� Obstacle avoidance� Sensor input

� Near real time

Page 76: Design Review

Planning

� Offensive position strategy

� Closest to ball goes; only 1 robot

� No passing; get ball, control till open shot

(Voronoi)(Voronoi)

� Defensive position strategy

� Block shot path (but don't mask goalie)

� Robot lay down in front of goal

� Determine offense/defense strat by which team is closest to ball

Page 77: Design Review

Communication

• Use of DGPS(Differential Global

Positioning)

• One fixed Satellite (Goalie Robot) and the

other two are Floating Satellites. The other two are Floating Satellites. The

Goalie robot will be doing all the analysis

and controlling the two robots.

• Sync the Transmission and Reception

signals between the Goalie and the other

two players.

Page 78: Design Review

Communication II

• Triangulation Method , easily used

as there are only 3 robots

(corresponding to three points on

the field)the field)

• Use of Ultra-Sonic Sensor, and by

determination of Time of Flight ,

positions can be determined by

analysis of the received reflected

Wave Packets

Page 79: Design Review

Communication III

• Special Beacon emitting a particular frequency employed on the two team robots, as to not get confused with the inter-mingled echo received when two opposition robots are close by and hence can be differentiated.close by and hence can be differentiated.

• Doppler radar to be used to understand the overall relative motion in the field, but the difference here is, the transmitter and the receiver is fixed.