Soccer without Reason Computer Vision and Control for Soccer Playing Robots Dr. Raul Rojas A new AI benchmark -computer vision in real time - embodied intelligence: mechanics - energy management - local control - communication between autonomous agents - team behavior - adaptation and learning
Robotic Soccer Started in 1997 RoboCup takes place together with IJCAI I - Simulation league II Small size league III- Mid-size league IV- Legged league V Humanoid league Simulation league soccer-server simple actions virtual robots
Legged league Mid-size league field 10 5 meters four on four
The mid-size environment Mid-size robots Omnidirectional vision - Laptop for control - Firewire video camera
Small size league 18 cm diameter Omnidirectional robots
I Global vision Global vision computer wireless communication global camera
The world is colored Team color Block diagram of the software user interface vision system reactive behavior wireless communication
Sensing colors Interpolate colors Illumination, aberration artifacts
Segmentation Original data high saturation hue Average color Tracking the robots Robot model 3 points, position, direction,... 4 frames! geometry :-) :-(
Tracking the ball the position of the ball is predicted P.-Vorhersage, RGB, HSI, Größe,... variable search frame color adaptation Adaptive color maps
Global Search RGB distance Subsampling Structure of the Vision System framegrabber: camera image Ball-Module Update- Module Team-Module Team-Module Field, Ball, Robots
II - Hardware Kicking device Motors Batteries Chassis Electronics Chassis Stable box for all devices
Electronics Wireless module PID controller HC12 Microprocessor From Motorola
Wireless modules Transmitter TX3 up to 64 kb/s Operation from 2.2V to 13V @ 7.5mA Receiver RX3 Operation from 2.7V to 10V @ 9.5mA Kicking device Rotation
II Local vision Our first omnivision robots
Spherical and parabolic transformations Hyperbolic and elliptic mirrors Focus on a pinhole
The field seen with our mirror Locating the robot Parabolic Mirror 500 400 distance 300 200 100 0 0 0.1 0.2 0.3 0.4 pixel distance
Distances to the sides 1 0.8 0.6 0.4 0.2 0-0.2-0.4-0.6-0.8-1 1 26 51 76 Series1 Series2 Series3 Series4 60 40 20 0-100 -80-60 -40-20 -20 0 20 40 60-40 -60-80 Series1 a) Expectation-Maximization 80 60 40 20 0-100 -50-20 0 50 100-40 -60-80 80 60 40 20 0-100 -50-20 0 50 100-40 -60-80 80 60 40 20 0-100 -50-20 0 50 100-40 -60-80 Series1 Series1 Series1 - Every line attracts the points nearer to it - Compute total force - Iterate 80 60 40 20 0-100 -50-20 0 50 100-40 -60-80 Series1
b) Locating a robot with two angles to known beacons III Reactive Behavior
Reactive Behavior Control slow fast sensors behaviors actuators Team Control Team Levels Robot 1 Robot 2 Robot 3 Robot 4 Robot 5
Sensor Aggregation: Robot position Layer 0: noise Layer 1: smoothed Subsampling Layer i+1 Layer i Layer i-1
Structure of a layer Higher layer sensors effectors sensors behaviors actors Lower layer Deciding on a kick Ball is on the other side Team: twice per second offense should_i = 0 should_i = 0 should_i = 1 should_i = 0 player 1 player 2 player 3 player 4 Homing=>covering Homing =>covering kick Homing=>keep free
player 3 Kick: at the level of the player Layer 2: every 16 Frames kicking Kick direction run finalizing Layer 1: every 4 Frames position Drive_for_run Rotate for run drive_for_finalizing rotate_for finalizing Layer 0: each Frame drive rotate Kicking reflex Kicking reflex activated
Ball prediction Screenshot of control software
Kicking Path planning objective desired position position approach Path planning stop execute Kicking aproach objective position Ziel approach Path planning stop execute
Kicking objective aproach position Ziel approach Path planning stop execute Kicking objective execute position approach Path planning stop execute
Kicking objective execute Halten approach Path planning stop execute Kicking objective execute stop approach Path planning stop execute
Kicking stop approach Path planning stop execute Pass approach goal goal position approach position
Pass Pass goal position goal aproach position pass Passen pass goal kick-it goal approach position pass kick-it
Pass kick-it approach position pass kick-it Pass kick-it approach position pass kick-it
Pass kick-it approach position pass kick-it Taxis
Goalie behaviors Midsize robots
Omnidirectional vision Die FU-Fighters
Tracking soccer players