STEERING BEHAVIORS Markt a Popelov, market a.popelova [ zavin ] mat - - PowerPoint PPT Presentation

Markét a Popelová, market a.popelova [zavináč] mat fyz.cz 2013, Human-like Art ificial Agent s, MFF UK Jakub Gemrot , gemrot @ gamedev.cuni.cz 2015, Human-like Art ificial Agent s, MFF UK

STEERING BEHAVIORS

MOTIVATION

Red S ea, Fish S chool https:/ / www.youtube.com/ watch? v=PLkpfKiinTA

MOTIVATION

Boid Particles S imulation, Blender https:/ / www.youtube.com/ watch? v=rLBrK0K-Ny8

MOTIVATION

BrainBugz https:/ / www.youtube.com/ watch? v=uRTo9qECQrM

REQUIREMENTS FOR MOTION CONTROL

• Responding t o dynamic environment
• Avoiding obst acles and ot her agent s
• Int eract ion wit h environment and ot her agent s
• Mot ion believabilit y
• Speed of comput at ion
•  One possible solut ion: St eering Behaviors by Craig W. Reynolds
• 1986 Flocks, Herds, and Schools: A Dist ribut ed Behavioral Model [1]
• Boids & Flocking Model
• 1999 St eering Behaviors For Aut onomous Charact ers [2]

STEERING BEHAVIORS - BASICS

• Hierarchy of mot ion behavior
• Act ion select ion layer + Pat h-finding
• St eering (navigat ion) layer
• Locomot ion layer
• Simple vehicle model
• Scalars: mass, max_force, max_speed
• Vect ors: locat ion, velocit y, orient at ion
• One st eering force:
• accelerat ion = st eering_force / mass  t runcat ed by max_force
• new_velocit y = original_velocit y + accelerat ion  t runcat ed by max_speed
• new_locat ion = original_locat ion + new_velocit y
• Or we can use f orces as direct input s f or physical simulat or

Locomotion

Animation, articulation

S teering

Path determination

Action S election

S trategy, goals, path-planning

BOIDS & FLOCKING MODEL

• Boid (bird like obj ect )
• Flocking Model  3 st eering rules

Separat ion

• Do not get t oo close

t o nearby flockmat es

• St eers boid from

t oo close flockmat es Alignment

• Try t o move at t he

same speed and direct ion (velocit y) as nearby flockmat es

• St eers boid t o have

t he same velocit y as t he average of velocit ies of nearby flockmat es Cohesion

• Prefer t o be at t he

cent er of t he local flockmat es

• St eers agent t o t he

cent er of nearby flockmat es

For figures and video see Craig Reynolds‘ web site http:/ / www.red3d.com/ cwr/ boids/ [3]

FLOCKING DEMONSTRATION I.

Interactive Boids

http:/ / blog.soulwire.co.uk/ laboratory/ flash/ as3-flocking- steering-behaviors

FLOCKING MODEL - FEATURES

• Relat ively believable
• Relat ively fast
• St raight forward implement at ion  O(n2)
• Using spat ial dat a st ruct ure for nearby flockmat es det ect ion  O(n)
•  Used in films and games
• E.g., Bat man Ret urns

FLOCKING DEMONSTRATION II.

S tanley & S tella in: Breaking the Ice http:/ / www.youtube.com/ watch? v=3bTqWsVqyzE

1999 C. REYNOLDS: STEERING BEHAVIORS FOR AUTONOMOUS AGENTS

• Seek & Flee
• Pursue & Evade
• Arrival
• Wander
• Obst acle Avoidance & Cont ainment
• Collision Avoidance & Unaligned collision avoidance
• Wall Following
• Pat h Following
• Leader Following
• Flow Field Following

SEEK & FLEE

• Seek
• st eers agent t o a st at ic t arget
• Flee
• st eers agent from a st at ic t arget

S eek steering force computation to_target = target_position – my_positin desired_velocity = normalize( to_target) * max_speed steering = desired_velocity – velocity

For figures and video see Craig Reynolds‘ web site http:/ / www.red3d.com/ cwr/ steer/ S eekFlee.ht ml [3]

ARRIVAL

Arrival steering force computation to_target = target_position – my_position distance = length( to_target ) ramped_speed = max_speed * ( distance / slowing_distance ) clipped_speed = min( ramped_speed, max_speed ) desired_velocity = to_target * ( clipped_speed / distance ) steering_force = desired_velocity – velocity

For figures and video see Craig Reynolds‘ web site http:/ / www.red3d.com/ cwr/ steer/ Arrival.ht ml [3]

• As Seek, except t he agent slows down as it approaches t he t arget

PURSUE & EVADE

• As seek & flee, except t he t arget moves
• Agent predict s t he locat ion of t he t arget in t he next t ick of t he

simulat ion

• Predict ion based on dist ance
• Nearer t he t arget is,

less t he predict ion (T) is used

For figures and video see Craig Reynolds‘ web site http:/ / www.red3d.com/ cwr/ steer/ PursueEvade.html [3]

WANDER

• Type of random st eering: t he st eering direct ion on one frame is relat ed

t o t he st eering direct ion on t he next frame

• More believable t han t ot ally random st eering forces
• St eering force:
• At each t ime st ep a random offset is added t o t he wander direct ion
• The modified wander direct ion is const rained t o lie on t he big circle
• Const rict ion of t he st eering: big circle
• Const rict ion of t he offset : small circle

For figures and video see Craig Reynolds‘ web site http:/ / www.red3d.com/ cwr/ steer/ Wander.ht ml [3]

WANDER

• St eering force:
• At each t ime st ep a random offset is added t o t he wander direct ion
• The modified wander direct ion is const rained t o lie on t he big circle

For figures and video see Craig Reynolds‘ web site http:/ / www.red3d.com/ cwr/ steer/ Wander.ht ml [3]

UNALIGNED COLLISION AVOIDANCE

• Separat ion
• Agent is st eered from t oo close neighbors
• Unaligned collision avoidance
• Pot ent ial collisions wit h ot her agent s are predict ed
• Agent is st eered t o avoid t he sit e of t he predict ed collision

For figures and video see Craig Reynolds‘ web site http:/ / www.red3d.com/ cwr/ steer/ Unaligned.html [3]

UNALIGNED COLLISION AVOIDANCE

For figures and video see Craig Reynolds‘ web site http:/ / www.red3d.com/ cwr/ steer/ Unaligned.html [3]

• Seeing head
• Defined by const ant
• Larger t he const ant
• Sooner an obst acle is det ect ed

UNALIGNED COLLISION AVOIDANCE

• Det ect ion by simple relat ion of t he dist ance t o an obst acle cent er
• Only closest t hread is considered
• And t he avoidance force generat ed

For figures and video see Craig Reynolds‘ web site http:/ / www.red3d.com/ cwr/ steer/ Unaligned.html [3]

UNALIGNED COLLISION AVOIDANCE

For figures and video see Craig Reynolds‘ web site http:/ / www.red3d.com/ cwr/ steer/ Unaligned.html [3]

• Seeing head
• Defined by const ant
• Larger t he const ant
• Sooner an obst acle is det ect ed

LEADER FOLLOWING

• Agent is st eered t o follow a Leader (grey).
• St eering force consist s of:
• Arrival – t he t arget is slight ly behind leader
• Separat ion – t o prevent collisions wit h ot her followers
• If a follower finds it self in a rect angular region in front of t he leader,

it will st eer lat erally away from t he leader’s pat h

For figures and video see Craig Reynolds‘ web site http:/ / www.red3d.com/ cwr/ steer/ LeaderFollow.html [3]

OBSTACLE AVOIDANCE

• Obst acle det ect ion
• Navigat ion graph, navigat ion mesh, et c.
• Point cont ent
• Line t races
• …
• Obst acle Avoidance by C. Reynolds
• An imaginary cylinder in front of t he agent should be free
• If it is free, t he st eering force is zero vect or
• Ot herwise it is t he vect or from t he most t hreat ening obst acle

• The most t hreat ening obst acle

is det ect ed and t he agent is st eered from it

• The agent ‘ s fut ure posit ion is

predict ed and t he agent is st eered t owards t he allowed region

OBSTACLE AVOIDANCE & CONTAINMENT

For figures and video see Craig Reynolds‘ web site http:/ / www.red3d.com/ cwr/ steer/ Obstacle.html [3] http:/ / www.red3d.com/ cwr/ steer/ Containment.ht ml [3]

OBSTACLE AVOIDANCE

(DYNAMIC) OBSTACLE AVOIDANCE EXAMPLE

Hitman Blood Money, 2009 http:/ / www.youtube.com/ watch? v=ycDi7fK797U

PROBLEMS AND DISCUSSION OF OA USE

• Problems wit h obst acles det ect ion
• Narrow obst acles
• Obst acles may not be det ect ed
• Obst acles may be det ect ed wrongly
• Ray lengt h
• Quick react ions vs. narrow passages
• Simulat ion frequency
• Specific sit uat ions
• Front collisions
• Local t raps and complicat ed sit uat ions
• OA uses only local informat ion

OBSTACLE AVOIDANCE & LOCAL TRAPS

OBSTACLE AVOIDANCE & LOCAL TRAPS

OBSTACLE AVOIDANCE & LOCAL TRAPS

WALL FOLLOWING

• Agent is st eered t o move in parallel wit h a wall
• The fut ure agent ‘ s posit ion is predict ed (t he black dot )
• This fut ure posit ion is proj ect ed t o t he nearest point on a wall (red dot )
• Red line represent s t he wall’s normal and leads t o t he t arget point (red

circle)

• Seek behavior is used t o st eer agent t owards t he t arget point
• Surface prot ocol:
• t he nearest point on t he wall
• t he normal at t hat point

For figures and video see Craig Reynolds‘ web site http:/ / www.red3d.com/ cwr/ steer/ Wall.ht ml [3]

WALL FOLLOWING

• Not es on possible implement at ion:
• Wall is det ect ed by rays
• At t ract ive force t o wall
• The fart her from wall an agent is, t he bigger t he at t ract ive force is
• Repulsive force from wall – if t he agent is t oo close t o wall
• The closer t o wall t he agent is, t he bigger is t he repulsive force
• Special sit uat ions
• Edges
• Front collisions

WALL FOLLOWING – DEMO I.

WALL FOLLOWING DEMO II.

WALL FOLLOWING IN COMBINATION

TA + OA

 direct ly t hrough cit y

TA + WF

 on t he sidewalks

FLOW FIELD FOLLOWING

• Flow field defines mapping: locat ion  flow vect or
• May be defined procedurally / based on dat a
• May be st at ic / t ime-varying
• The fut ure locat ion is predict ed
• F = flow vect or at t his locat ion
• st eering_force = velocit y – F

For figures and video see Craig Reynolds‘ web site http:/ / www.red3d.com/ cwr/ steer/ FlowFollow.html [3] Vector-field Pathfinding https:/ / www.youtube.com/ watch? v=Bspb9g9nTto

S upreme Commander 2, 2010 http:/ / www.youtube.com/ watch? v=j A2epda-RkM

COMBINING STEERING BEHAVIORS

• Each st eering behavior ret urns single vect or

(st eering force)

• What t o do wit h more st eering behaviors?
• Select and apply t he most import ant st eering behavior
• Select random act ive st eering behavior
• Sum all forces t oget her
•  Average of all forces
•  Average of all non-zero forces
•  Weight ed average of all non-zero forces

STEERING BEHAVIORS FOR IVA‘ S

• Which mot ion problems do we deal wit h in applicat ions wit h IVA‘ s?
• Where would be st eering behaviors helpful?

REAL EXAMPLE I.

No collision avoidance S tronghold Crusader Extreme, 2008 http:/ / www.youtube.com/ watch? v=IZpgMnu_lAk

REAL EXAMPLE II.

S mall collision radius Dawn of War, 2009 http:/ / www.youtube.com/ watch? v=IZpgMnu_lAk

REAL EXAMPLE III.

Primitive (and slow) collision avoidance Knights and Merchants, 1998 http:/ / www.youtube.com/ watch? v=IZpgMnu_lAk

REAL EXAMPLE IV .

Getting stuck Empire Total War, 2009 http:/ / www.youtube.com/ watch? v=IZpgMnu_lAk

PROBLEMS OF LOCAL INFORMATION

• Complicat ed t asks can not be solved
• What t o do?
•  use global knowledge of t he environment
•  plan t he pat h

DIRECT FOLLOWING OF THE PLANNED P ATH

PROBLEMS OF DIRECT FOLLOWING

• Not believable
• Somet imes lacks smoot hness
• What t o do?
•  st eering behavior Pat h Following
• Paramet ers: pat h (a list of locat ions), dist ance from pat h
• Not es on implement at ion
• Pair of pat h nodes
• Force t o t he cent er axis
• Improvement s
• Proj ect ion lengt h
• Regulat ion force

P ATH FOLLOWING

• Agent is st eered t o move along t he pat h in t he given direct ion while

keeping it s cent er in t he gray region

For figures and video see Craig Reynolds‘ web site http:/ / www.red3d.com/ cwr/ steer/ PathFollow.html [3] http:/ / www.red3d.com/ cwr/ steer/ CrowdPath.ht ml [3] GameDev Tutorials http:/ / gamedevelopment.t utsplus.com/ tutorials/ understanding-steering-behaviors-path-following--gamedev-8769

PEOPLE (COLLISION) AVOIDANCE

• Basics
• Repulsive force from ot her t oo close agent s
• Problems

 Circumvention

(rotational force)

 Acceleration &

deceleration 1 2

PEOPLE AVOIDANCE – DEMONSTRATION

OTHER SOCIAL INTERACTIONS

• Leader Following
• Walk Along [10]
• Two friends go t oget her t o a cert ain place
• Ot her…

?

WALK ALONG - DEMONSTRATION

ACTION SELECTION LAYER

• Which st eering behavior should be act ive?
• Paramet ers?
• Should be cont rolled by act ion select ion layer
• Aut onomously vs. Cent rally
• Some problems could be solved on t he act ion select ion layer
• Pat h Following vs. Ot hers
• Commander and his regiment
• Det ect ion of being st uck, et c.
• Set t ing paramet ers according t o mood, emot ions et c.

Locomotion

Animation, articulation

S teering

Path determination

Action S election

S trategy, goals, planning

STEERING BEHAVIORS CONCLUSION

• Advant ages
• Simplicit y  predict abilit y (good for debugging)
• React ive behavior  efficiency (t ime, memory)
• Forces  smoot hness, combinabilit y
• Disadvant ages
• Simplicit y & Local Traps  low believabilit y  somet imes we need

higher-level predict ion and planning

• Scalabilit y (modifying t he behavior by hacking ext ra lines int o code)
• Use
• Comput er games, Films
• Crowd simulat ions (evacuat ions, shopping cent ers, et c.)

WORKSHOP

• St eeringTool 2.0
• St eeringGame
• UT2004St eeringLibrary

Web & Instalator: http:/ / diana.ms.mff.cuni.cz/ pogamut-games Proj ect S teeringGame

LITERATURE I.

1. REYNOLDS, Craig W. Flocks, Herds, and Schools: A Dist ribut ed Behavioral

• Model. In Proceedings of Comput er Graphics. Anaheim, California : ACM

SIGGRAPH, 1987. Pages 25-34. WWW: <ht t p:/ / www.red3d.com/ cwr/ papers/ 1987/ SIGGRAPH87.pdf>. 2. REYNOLDS, Craig W. St eering Behaviors For Aut onomous Charact ers. In Proceedings of Game Developers Conference. San Francisco, California : Miller Freeman Game Group, 1999. Pages 763-782. WWW: <ht t p:/ / www.red3d.com/ cwr/ papers/ 1999/ gdc99st eer.pdf>. 3. REYNOLDS, Craig W. St eering Behaviors For Aut onomous Charact ers [online]. Sept ember 5, 1997 , June 6, 2004 [cit . 2011-05-19]. St eering Behaviors For Aut onomous Charact ers. WWW: <ht t p:/ / www.red3d.com/ cwr/ st eer>.

Basics, Craig Reynolds, Boids, and

• riginal S

teering Behaviors

LITERATURE II.

4. CHAMP ANDARD, Alex J. AI Game Development : Synt het ic Creat ures wit h Learning and React ive Behaviors. First print ing. Unit ed St at es of America : New Riders Publishing, 2003. ISBN 1-5927-3004-3. 5. CHAMP ANDARD, Alex J. AI Game Programming Wisdom 2. First Edit ion. Unit ed St at es of America : Charles River Media, 2004. An Overview of Navigat ion Syst em, Pages 131-139. ISBN 1-58450-289-4. 6. SINGH, Shawn, et al. Wat ch Out ! A Framework for Evaluat ing St eering

• Behaviors. In Mot ion in Games : First Int ernat ional Workshop, MIG 2008

Ut recht , The Net herlands, 2008 Revised Papers. Germany : Springer- Verlag, 2008. Pages 200-209. ISSN 0302-9743. 7. KARAMOUZAS, Ioannis, et al. A Predict ive Collision Avoidance Model for Pedest rian Simulat ion. In Mot ion in Games : Second Int ernat ional Workshop, MIG 2009 Zeist , The Net herlands, 2009 Proceedings. Germany : Springer-Verlag, 2009. Pages 41-52. ISSN 1867-8211.

Related works, Benchmark for S teering Behaviors, Collision Avoidance Model

LITERATURE III.

8. POPELOVÁ, Markét a; BÍDA, Michal. St eering t echinky pro virt uální agent y. In KELEMEN, Jozef; KVASNIČKA, Vladimír; POSPÍCHAL, Jiří.

Kognice a umělý život XI. Opava : Slezská univerzita v Opavy, 2011. Pages

207-212. ISBN 978-80-7248-644-1. 9. POPELOVÁ, Markét a. Knihovna st eering t echnik pro virt uální agent y. Bachelor t hesis. Charles Universit y in Prague, 2011. WWW: <ht t p:/ / amis.mff.cuni.cz/ emohawk/ > (8.1.2012).

• 10. POPELOVÁ, Markét a, et al. When a Couple Goes Toget her: Walk Along

St eering. In Proceedings of Mot ion in Games, Lect ure Not es in Comput er

• Science. Volume: 7060, Springer, Heidelberg, Pages 278-289, ISBN 978-3-

642-25089-7, 2011.

S teering Behaviors for IVA‘ s, S teering Behavior with social aspect