Hierarchical Model for Real Time Simulation of Virtual Human Crowds

1
Hierarchical Model for Real Time Simulationof
Virtual Human Crowds

• Soraia Raupp Musse and Daniel Thalmann

2
Contents

• Introduction
• Related work
• Crowd structure
• Crowd information
• ViCrowd architecture
• Behaviors
• Results discussion
• Conclusions

3
1. Introduction (1/3)

• ViCrowd
• A behavioral-based multi-level framework to
  simulate virtual human crowds in real time
• Concepts
• Virtual human agent
• Simulation Entity Crowd Groups Agents
• crowd set of groups, group a group of agents
• Intentions, beliefs, knowledge
• Crowd behavior
• Events

4
1. Introduction (2/3)

• Problems
• Modeling of crowd information and hierarchical
  structure, also concerning its distribution among
  groups
• Different levels of realism variation in
  behavior complexity
• The required structure to provide interaction
  with groups of agents during the simulation in
  real-time
• Solutions
• A hierarchy composed of crowd, groups, and agents
  is used? Crowd structure
• Endowing virtual agents with different levels of
  autonomy ? Crowd behavior (programmed, reactive,
  autonomous, guided)

5
1. Introduction (3/3)

• Group based model
• Automatic generation of human crowds based on
  groups
• Groups are more intelligent than individuals.
• Real time Requirements and information
  optimization
• Based on flocking systems
• Contributions
• Multi-level hierarchy
• Various degrees of autonomy
• Group-based behaviors

6
2. Related Work (1/2)

• Modeling the motion of groups with significant
  physics
• Particle systems Bouvier
• Crowd simulation with dynamics Brogan
• Distributed behavior model for simulating flocks
  Reynolds
• Methods to simulate the movement of pedestrians
  Helbing
• Local rules for controlling collective behaviors
  Matraric, Noser
• Behavioral animation for creating artificial life
  Terzopoulos
• Presenting the problem of building autonomous
  animated creatures for interactive virtual
  environments Blumberg
• IMPROV a system for scripting interactive actors
  Perlin
• Simulating crowds of ants in an automatic way
  PDI, PIXAR
• Bugs Life 4466 different individual motions to
  describe 228 different behaviors

7
2. Related Work (2/2)
8
3. Crowd Structure (1/2)

• Distributing the crowd behaviors to the groups
  and then to the individuals

9
3. Crowd Structure (2/2)

• Two ways of setting the parameters
• Scripted control defines scripted behaviors of
  the crowd
• External control specifies guided behaviors
• The intelligence, memory, intention and
  perception are focalized in the group structure.

10
4. Crowd Information (1/17)

• Knowledge/belief/intention

11
4. Crowd Information (2/17)

• Knowledge
• Information of the virtual environment
• the real position of a chair
• concerns the memory of groups related to the past
  experiences as well as perception related to
  agents and groups.
• Crowd obstacles
• Crowd motion and action
• Group Knowledge

12
4. Crowd Information (3/17)

• Knowledge
• Crowd obstacles
• declaration of all objects of the scene
• declaration of the areas where the crowd can
  walk.
• declaring some regions where the crowd can walk
  with some obstacles to be avoided

13
4. Crowd Information (4/17)

• Knowledge
• Crowd motion and action
• Described using goals
• Goals IP AP
• Individual AP
• Shared AP
• The agents from the same group share the same
  goals.
• The paths for the different agents from the same
  group can be similar but are never the same
  because they cannot occupy the same sub-region.

14
4. Crowd Information (5/17)

• Knowledge
• Group knowledge
• Memory
• A structure where the leaders perceived
  information can be stored and processed
  afterwards depending on the specified behavioral
  rules ? The memory of groups is processed only
  by the leader of the group
• The size of the memory (capacity of storage) can
  be pre-defined for each group of crowd.
• Group perception
• Some group associated parameters knowledge,
  beliefs and intentions
• Location of groups/agents
• Associated to just the leader of group

15
4. Crowd Information (6/17)

• Beliefs
• Emotion the list of behaviors to be applied by
  the groups
• Can be shared or be redefined.
• Crowd and group behaviors
• High-level behaviors for crowds programmed in the
  script language or directly informed using guided
  control
• 8 group behaviors
• Flocking, following, goal changing, attraction,
  repulsion, split, space adaptability,
  safe-wandering

16
4. Crowd Information (7/17)

• Beliefs
• Crowd and group behaviors
• Flocking
• Group ability to walk together at the same speed
  towards the same goals
• In real life, people walk in groups
• Rules for flocking
• The agents from the same group share the same
  list of goals
• They walk at similar speeds
• They follow the paths generated in Section 3.2.1
• One agent can wait for another on arrival at a
  goal when another agent from the same group is
  missing

17
5. Crowd Information (8/17)

• Beliefs
• Crowd and group behaviors
• Following
• Group ability to follow a group or an individual
  motion
• Permanent/temporary
• Goal changing
• Agents can have the intention of changing groups
• The relationship with all groups
• Its domination status

18
4. Crowd Information (9/17)

• Beliefs
• Crowd and group behaviors
• Attraction
• Groups of agents are attracted around an
  attraction point

19
4. Crowd Information (10/17)

• Beliefs
• Crowd and group behaviors
• Repulsion
• Groups ability to be repulsed from a specific
  location or region

20
4. Crowd Information (11/17)

• Beliefs
• Crowd and group behaviors
• Split
• The subdivision of a group to generate one or
  more groups

21
4. Crowd Information (12/17)

• Beliefs
• Crowd and group behaviors
• Space Adaptability
• Group ability to occupy all the walking space

22
4. Crowd Information (13/17)

• Beliefs
• Crowd and group behaviors
• Safe-Wandering
• Evaluate and avoid collision contacts with agents
  and objects

23
4. Crowd Information (14/17)

• Beliefs
• Emotional status
• Subjective climate to be simulated
• E.g. sad, calm, regular, happy,
  explosive
• Parameters Way of walking, Walking speed, The
  range of basic actions
• Emotion distribution
• Emotional status can trigger an event.
• Emotion can be changed as a function of triggered
  events and reactions

24
4. Crowd Information (15/17)

• Beliefs
• Individual beliefs
• Basically, individuals are just able to walk
  whereas avoid collision with obstacles and other
  agents.
• When the sociological effects are applied, its
  possible for the individuals to have a more
  complex structure of parameters including
• Goal-changing behavior, group behavior (see
  Section 3.2.2.1)
• A value for the relationship with all groups
  (between 0 and 1)
• A value for its domination status

25
4. Crowd Information (16/17)

• Intentions
• Groups intentions
• Goals and actions to be applied
• Individuals intentions
• Follow group or change of groups
• Be leader dependent on domination value

26
4. Crowd Information (17/17)

• Inter-Relationship among information

27
5. ViCrowd Architecture

• Low-level behavior generation for each
  group
• Modification of low-level behaviors of
  groups due to sociological effects
• Modification of low-level behaviors of
  groups due to reaction triggering events
  (informed in the script language or by the
  external control)
• Distribution of groups low-level behaviors
  to the individuals

28
6. Behaviors (1/8)

• Programmed behaviors
• Script language
• To program specific behaviors in ViCrowd
• To deal with three kinds of information
• Geometric information
• Behavioral information
• Simulation information
• To provide predefined commands to easily specify
  crowd behaviors

29
6. Behaviors (2/8)

• Programmed behaviors
• IPs and APs Generation

IP0 POSITION ( 1000 0 2500 ) ORIENTATION ( 0.87 0
0.4 ) ASSOCIATED_REGION 3 PTS ( 1300 0 5600 ) (
2100 0 4200 ) ( 1250 0 200 ) AP0 POSITION ( 1000
0 2500 ) ORIENTATION ( 0.87 0 0.4
) ASSOCIATED_REGION 2 PTS ( 1300 0 5600 ) ( 2100
0 4200 ) TYPE SHARED ACTION PARAMETERS KEYFRAME
applause
30
6. Behaviors (3/8)

• Programmed behaviors
• Crowd behavior generation

CROWD_SEMANTIC INTENTION PERCENTAGE 20 GOES TO
IP0 APPLIES AP0 PERCENTAGE 40 DOES
NOTHING BELIEFS EMOTION ALL HAPPY LEADER
PERCENTAGE 10 GROUP_BEHAVIORS ADAPTABILITY
PERCENTAGE 20 FLOCKING GROUP_5 SET
SOCIOLOGICAL_EFFECTS ON
31
6. Behaviors (4/8)

• Reactive behaviors
• Autonomous crowd
• Crowd has the ability to react to events.
• Event
• can trigger a reaction.
• Global events, local events, emotional events
• ltwhengt, ltwhogt
• Reaction
• Contains the definition of the behaviors to be
  applied.
• ltmotiongt, ltactiongt, ltinternal_status_changegt,
  ltposturegt, etc

32
6. Behaviors (5/8)

• Reactive behaviors
• Event
• Reaction

event BUS_STOP WHEN CONDITION (Object_bus NEAR
IP_BUS) WHO ALL_GROUPS NEAR IP_BUS
Reaction TAKE_THE__BUS ACTION KEYFRAME
take_bus MOTION ATTACH Object_bus CHANGE_EMOTI
ON HAPPY
33
6. Behaviors (6/8)

• Reactive behaviors

CROWD_STATUS PARTY PERCENTAGE 70
HUNGRY PERCENTAGE 30 SOCIAL
event HUNGRY WHO GROUP_STATUS HUNGRY WHEN
EVENT FOOD MATCHED reaction HUNGRY MOTION GO TO
TABLE / TABLE IP / ACTION INTERACTION SOBJ
TABLE
event CHANGE WHO GROUP_STATUS HUNGRY AND WHEN
INTERACTION SOBJ TABLE MATCHED AND REACTION
HUNGRY MATCHED reaction CHANGE GROUP_STATUS
SOCIAL
34
6. Behaviors (7/8)

• External behaviors
• Guided crowd
• A client/server system to combine a Rule-Based
  Behavior System (RBBS) and ViCrowd
• The RBBS is able to control groups or individuals
  by sending the information specifying what
  motion, action, event or reaction is to be
  applied.
• Behavioral rules have a simple syntax

rule2 if ((Laura needs to take the train and
(Laura has no ticket)) then (Laura goes to
counter to buy one)
35
6. Behaviors (8/8)

• External behaviors
• Group memory
• A guided behavior has priority over all scripted
  behaviors.

36
7. Results Discussion (1/2)

• Simulations of crowds can be done in general
  virtual environments

(SB)(GB) (RB)
(SB)
(SB)(RB)
(SB)(GB)
37
7. Results Discussion (2/2)
38
8. Conclusions

• The possibility of generating simulations with
  various levels of realism, including scripted,
  reactive, and guided behavior.
• The hierarchical structure used to model crowds
  based mainly on groups instead of individuals.