2012 simulated car racing championship @ cig-2012

2012 Simulated Car Racing Championship @ CIG-2012

The 2012 Simulated Car Racing Championship @ CIG-2012 Daniele Loiacono, Luigi Cardamone, and Pier Luca Lanzi


2012 Simulated Car Racing Championship 9 races during 3 conferences

Develop a driver for TORCS (hand-coded, learned, evolved, …)

Drivers will be awarded based on their score in each conference competition

At the end, the team with highest overall score wins the championship


Roadmap of the 2012 Championship

1.  EVO*-2012, Malaga (Spain) April 11-13, 2012

2.  ACM GECCO-2012, Philadelphia (US) July 7-11, 2012

3.  IEEE CIG-2012, Granada (Spain) September 11-14, 2012


Motivations

q  Proposing a relevant game-based competition " more representative of commercial games AI " more similar to a real-world problem

q  Proposing a funny and exciting competition " you can see and play with the entries of this competition " human players can interact with AI " a lot of programmed AI available for comparison

q  Proposing a challenging competition " computationally expensive " real-time " on-line learning " noisy sensors

The Open Racing Car Simulator


The Open Racing Car Simulator & the Competition Software

TORCS

BOT BOT BOT

TORCS PATCH

SBOT SBOT SBOT

BOT BOT BOT

UDP UDP UDP q  The competition server

q  Separates the bots from TORCS q  Build a well-defined sensor model q  Works in real-time


Sensors and actuators

q  Rangefinders for edges on the track and opponents (with noise) q  Speed, RPM, fuel, damage, angle with track, distance race, position

on track, etc.

q  Six effectors: steering wheel [-1,+1], gas pedal [0, +1], brake pedal [0,+1], gearbox {-1,0,1,2,3,4,5,6}, clutch [0,+1], focus direction


What is the structure of a race?

Three stages: warm up, qualifiers, actual race

During warm-up, each driver can explore the track and learn something useful

During qualifiers, each driver races alone against the clock (the best 8 drivers move to the race)

During the race all the drivers race together

Competitors


The competitors

q  Four entries in 2012: " Ho Duc Thang, University of Nottingham " BFB4, Sejong Univ, Korea " Mr. Racer, TU Dortmund " Smile Sheep, Taiwan National Chiao Tung University

q  Three reference entries (selected from the top entries of the past championship) : " Mariscal & Fernández, Málaga University " Ready2Win, Slovak University of Technology – FIIT " AUTOPIA, Madrid and Granada


Ho Duc Thang University of Nottingham


Ho Duc Thang

q  Hybrid fuzzy controller with a module to learn the model track as the car drives, a module to perform some simple plannings based on the modelled segments of the track and a fuzzy controller which actually drives the car.

q  All the modelling, planning and driving are done during the race, the warm-up stage is not exploited

q  No opponent handling


Kyung-Joong Kim Jun-ho Seo Tae-seong Kim Sejong Univ, Korea

BFB4


BFB4

q  Scripted approach " Simple approach for steering and accelerating

q  Jump Detection " When the Z-Sensor is more than 0.4 " While the car is jumping and landing(few times, 10 tick

count), the steer of car is fixed to forward

q  Opponent avoidance: " Just brake if the opponent is approached on a straight " Modify the normal steering when the opponent is

approached in a bend


Jan Quadflieg, Tim Delbruegger and Mike Preuss TU Dortmund

Mr Racer


Mr. Racer 2012

q  Joined effort of Jan Quadflieg, Tim Delbruegger and Mike Preuss

q  Variables learned offline with the CMA-ES, now including optimized settings for recovery , the clutch and…

q  NEW: Parameters of the opponent handling module have been optimized

q  Optimization of the opponent handling will hopefuly result in a CIG competition paper


Noise Handling

1.  Low pass filter on sensor values

2.  2 Regression polynoms are fitted to each side

3.  Resample the polygons 4.  Calculate our measure (see

CIG paper 2010) 5.  Low pass filter on the last

measurements


Online Learning during Warmup

q  We learn a model of the racetrack q  One of two parameter sets is chosen q  Target speeds for all corners are then fine tuned q  All information is saved at the end of the warmup q  And later used during qualifying and the race


Opponent Handling

q  Noise filtering with a lowpass filter q  Observer module interprets filtered sensor readings q  This leads to a recommended speed and an overtaking line

(created on the fly!) q  Planning module incorporates recommended target speed

and overtaking line into the plan


吳晞浩(Si-Hao Wu) 李揚(Yang Lee) 李懷哲(Wiser Lee) Advisor: 王才沛(Tsaipei Wang) Taiwan National Chiao Tung University

Smile Sheep


Smile Sheep

q  Steering heuristics " drive toward the track sensor with biggest distance " small corection toward middle of the track to avoid

borders q  Speed control heuristics

" Full accelerating when central sensor is clear for >99m " Slow down when car is too fast

q  Recovery policy heuristics " identify stuck condition with heuristics " drive backward and get the car parallel to the track


Mariscal & Fernández

q  General approach: expert systems improved by multi-objectives evolutionary computation techniques " Maximize distance. " Minimize damage

q  Steering based on a simple rule follow the track sensor with biggest distance.

q  Overtaking based on simple rules and expert systems


Ready2Win

q  Modular architecture: " Driving module:

•  The speed is computed using the maximum braking distance

" Overtaking module " Recovery module " ABS module

q  Track learning during the warm-up: " First lap, driven slow, to identify turns (start, end, entry

position, curvature) and learn the track model " Other laps, speed adaptation:

•  Increasing according to lateral speed •  Reducing when the car goes off-road


Industrial Computer Science Department. Centro de Automática y Robótica Consejo Superior de Investigaciones Científicas Madrid, Spain Contact:E. Onieva ([email protected])

AUTOPIA


AUTOPIA

q  Fuzzy Architecture based on three basic modules for gear, steering and speed control " optimized with a genetic algorithm

q  Learning in the warm-up stage:

" Maintain a vector with as many real values as tracklength in meters.

" Vector initialized to 1.0 " If the vehicle goes out of the track or suffers damage

then multiply vector positions from 250 meters before the current position by 0.95.

q  During the race the vector is multiplied by F to make the driver more cautious in function of the damage: " F=1-0.02*round(damage/1000)

Qualifying


Scoring process: Warm-up & Qualifying

q  Scoring process involves three tracks: " IllSchwang (desert track) " Noceda (city track) " Sehnor (hill track)

q  The tracks are not distributed with TORCS: " Generated using the Interactive Track Generator for

TORCS and Speed Dreams available at: •  http://trackgen.pierlucalanzi.net

" The competitors cannot know the tracks q  Each controller raced for 100000 game ticks in the warm-up

stage and then its performance is computed in the qualifying stage as the distance covered within 10000 game ticks


Illschwang


Qualifying: Ilschwang

10891.4

10808.2

10701.6

10010.1

5066.24

6998.2

7765.42

0 2000 4000 6000 8000 10000 12000

Mr. Racer

Ho Duc Thang

Autopia

Ready2Win

Mariscal & Fernandez

BFB4

SmileSheep


Noceda


Qualifying: Noceda

12233.2

11738.82

11648.9

9405.52

11265.2

9658.8

0 2000 4000 6000 8000 10000 12000 14000

Mr. Racer

Ho Duc Thang

AUTOPIA

Ready2Win


BFB4


Sehnor


Qualifying: Sehnor

10483

10280.7

9646.59

9413.6

9610.96

9043.72

8683.94

0 2000 4000 6000 8000 10000 12000

Mr. Racer

Autopia

Ho

Ready2Win


BFB4

SmileSheep


Qualifying summary (scores with noise)

Driver Illschwang Noceda Sehnor Total

Mr. Racer 10 10 10 30 Ho Duc Thang 8 8 6 22 AUTOPIA 6 6 8 20 Ready2Win 5 3 4 12 Mariscal&Fernandez 2 5 5 12 BFB4 3 4 3 10 SmileSheep 4 2 2 8

CI Approaches


What about the qualifying results?

Since 2009 AUTOPIA is not the

fastest controller for the first time!

Top controllers clearly better than others


Three Tracks

For each track we run 6 races with different starting grids

Each race is scored using the F1 point system

(10 to first, 8 to second, 6 to third, …)

Two points to the controller with lesser damage

Two points for the fastest lap of the race


Final Results

Mr. Racer is winner of SCR@CIG 2012

Driver Illschwang Noceda Sehnor Total

Autopia 10 12 10 32 Mr. Racer 10 6 10 26 Ready2Win 8 3 6 17 SmileSheep 5 5 5 15 Mariscal&Fernandez 4 6 4 14 BFB4 3 5 4 12 Ho 2 2 2 6

What about the race results?

Qualifying and final results are similar: the only excepFon is driver Ho

Some drivers were not able to complete the

race

Further analysis will be performed to have more insight on these results!

CompeGtor Evo* Gecco CIG Total AUTOPIA 39 35 32 96 Mr. Racer 16.5 18 26 60.5 Ready2Win 19 20 17 56 Mariscal & Fernandez 20 18,5 14 52.5 BFB4 -‐ 11 12 23 Ho Duc Thang -‐ 13 6 19 Smile Sheep -‐ -‐ 15 15

Final Standings Championship

CompeGtor Evo* Gecco CIG Total AUTOPIA 39 35 32 96 Mr. Racer 16.5 18 26 60.5 Ready2Win 19 20 17 56 Mariscal & Fernandez 20 18,5 14 52.5 BFB4 -‐ 11 12 23 Ho Duc Thang -‐ 13 6 19 Smile Sheep -‐ -‐ 15 15

Final Standings Championship

Mr. Racer WINNER of 2012 Simulated Car

Racing Championship

We wait for you…

www.geccocompetitions.com

[email protected] @geccocomp /geccocompetitions

…next year!