Situation Based Approach for Virtual Crowd Simulation Ph.D Preliminary talk Mankyu Sung
Jan 02, 2016
Features in Crowds
• Large number of people• Share same environment• Anonymity• Importance of short term crowd behavior• Importance of locational factor in crowd behavior
• Importance of social-relational factor in crowd behavior
Why Crowd Simulation is Hard?
- Conflicting Goals- Simple agent with simple behaviors vs. Complex
agent with realistic behaviors
- Control over action of crowd vs. Not control over every agent individually
- Fast simulation of the small number of characters vs. Slow simulation of the large number of characters
The Goal of Research
• Set three demands that are able to solve these problems.
Scalability
Controllability
Convincingness
Scalability
• Two Specific Scalabilities– Memory Scalability
• The amount of memory for a character does not proportionally increase as the complexity of environment increases.
– Performance Scalability• The overall performance (frame-rate) does not
proportionally increase as the complexity of environment increases.
Convincingness
• Visually Convincing Behaviors– Visually realistic motion of characters
• Semantically Convincing Behaviors– Plausible behaviors for given time
• e.g.) At a crosswalk, crowds are crossing or standing depending on a traffic sign.
Controllability
• Specify crowd behaviors– User interfaces
• Control crowd flow– Predefined scenario– Interactive control– Density control
Proposed Approach
• Scalability– Situation based simulation
• Convincingness– STM(Snap-Together-Motion)– Composable behaviors
• Controllability– Painting interface– Situation graphs
(Sung et al. EG2004)
SituationBased
Approach
Thesis Statement
It is my thesis that the situation based approach is able to achieve the demand of
scalability, convincingness and controllability.
Related Works
• Smart Environments– Smart Object (Kallman et al. 1998)– Informed environment (Farenc et al. 1999)– Informed hierarchical information (Thomas et al. 200
0)– Apply Gibson’s “natural movement” theory (Michael et
al. 2003)
• Computer Games– The Sims TM (EA games)
Related Works
• Character Animation– Non-human creature
• Flocking algorithm : Boids (Reynolds, 1987)• Artificial fish by using synthetic vision (Tu et al. 1994)
– Human animation• Motion blending (Rose et al. 1996, Wiley et al. 1997,
Kovar et al. 2003)• STM(Snap-Together-Motion) (Gleicher et al. 2003)
Related works
A1 A2 A3
Actions
Time
• Behaviors in STM– Behavior is a series of actions over time– Specifying a behavior is to choose proper
action one by one in time
Related Works
• Intelligent Agent– Cognitive architecture (Funge 1999)
– Role-passing system (Horswill 1999, O’Sullivan et al. 2002, McNames et al. 2003)
• Crowd Modeling– Rule based system (Musse et al. 1987, 2001)– Cellular automata (Blue et al. 1998)
Related Works
• Crowd Modeling– Physically Based Approach
• Fluid dynamics (Henderson, 1974)• Particle system (Bouvier et al. 1997, Gipps et al. 1985)• Social force model (Helbing et al. 1995, 2000)
– Robotics Algorithm• Use PRM for group behavior (Bayazit et al. 2002)• Collision-free path planning for multiple robots (Furtney 200
0)• Leader-Following model (Li et al. 2001)
Situation Based Approach
• Scalability–Situation based simulation
• Convincingness– STM(Snap-Together-Motion)– Composable behaviors
• Controllability– Painting interface– Situation graphs
SituationsSituation
A1 A2
A3A8
Agent
Behavior 1
Behavior 2
Character
Actions
A1 A2
A3
A4
A8
A7
A6 A5
Behavior 1
Behavior 2
Behavior 3...
…
Situations (2)• Example
A man
Actions
sing walk
turn
sit
climb
dance
stand cross
Zig-Zag walk
Straight walk
Sit down...
…
At a crosswalk
A man
crossstreet
stand
Straight walk
Checking cars
Situation
Situations (3)
AgentA1
A2
Behavior 1Behavior 1
Behavior 2Behavior 2
A3
A4
Behavior 3Behavior 3
Behavior 4Behavior 4
Behavior 5Behavior 5
AugmentedBehaviors
AugmentedActions
PluggableAgent
Architecture
PluggableAgent
Architecture
Situation (4)
• Spatial Situation– Has a region in the environment
• e.g.) ATM, Bus Stop, Bench, Ticket Booth, Crosswalk
– The region is used for checking whether or not an agent is in the situation.
• Non-Spatial Situation– Social relationship between agents– Has no region in the environment– Directly set on crowds.
• e.g.) Friendship, Group member
Situation(5)
• Situation architecture
ActionsActions SensorsSensors
BehaviorFunctions
BehaviorFunctions
Event Rules
Event Rules
WalkWalk
TurnTurn
SitSit
Don’t’ turnDon’t’ turn
Don’t overlapDon’t overlap
Path planPath plan
Empty sensor
Empty sensor
Proximity sensor
Proximity sensor
Signalsensor
Signalsensor
If(Empty) then
Compose(Sitdown)
If(Signal) then
Compose(walk)
If(Empty) then
Compose(Sitdown)
If(Signal) then
Compose(walk)
Situation B
Situation(6)• Situation Composition
– Union of all components of situations
Situation A
Composed Situation
Situation C
Agent can react to the situation A, B and C at
the same time
Situation(7)• Example
Crossing to the other side of
The road
Traffic sign
Crossing a streetwith
checking traffic signs
Situation(8)
• Advantages of situation based simulation– Scalability
• Situation controls a small set of local behaviors.• Agents keep only information of the situations that
they are in at any given time.• Situations can be composed/decomposed easily.
– Ease of authoring– Re-usability– Efficiency
Situation Based Approach
• Scalability– Situation based simulation
• Convincingness– STM(Snap-Together-Motion)– Composable behaviors
• Controllability– Painting interface– Situation graphs
STM(Snap-Together-Motion)
• For visual convincingness, we use STM technique for animating characters.
– From input motion clips, the STM produces a set of small motions that can be connected with each other with minimizing artifacts.
[Gleicher et al. I3D 2003]
Composable Behaviors• For semantically convincing behaviors, we propo
se the composable behavior technique based on the probability scheme.
Agent
A1
A2
A3
Probability
Probability
Probability
Default Actions
Action from a situation
Actions
Composable Behaviors (2)
• Probability Scheme– Behavior functions compute the probability of
each action based on its own criteria.– Returned probability distributions are
composed by multiplication operation.– A sampling is performed on the final
probability distribution result to select a final action.
OverlapBehavior Function
Target FindingBehavior Function
.5 .5 .5
.3
.7.6
.19
.43.37
Composed Prob. Dist
Re-normalization
Actions
Actions
Actions
P(action)
P(action)
P(action)
Collision withOther agents
Agent has aTarget pos.
A B C
A B C
A B C
A B C
Multiplication
Composable Behaviors (4)
0 1
.43 (B).19 (A) .37 ( C)
Sampling (0-1)
.19
.43.37
Composed Prob. Dist
Actions
A B C
Action selection through sampling
Composable Behaviors (5)
• Advantages
– Gives a basic framework for scalability and controllability demands.
– Provides randomness on simulation– Takes various kinds of factors into account for
behaviors.
Situation Based Approach
• Scalability– Situation based simulation
• Convincingness– STM(Snap-Together-Motion)– Composable behaviors
• Controllability–Painting interface–Situation graphs (future work)
Painting Interfaces• How to specify a
particular situation in the environment.
SpatialSituation
Non-spatialSituation
Putting Pieces Together
Preprocessing
Create an environment
Set situations
Put crowds in the environment
Simulation time
Set run time situations
Situations
Plug-in information to agents
Checking events with sensors
Behavior composition
Sampling on final prob. dist
Demos
• 1. Composable behaviors
• 2. Street environment
• 3. Theater environment
• 4. On-line situation setting
• 5. Painting interface
• 6. Visualization of crowds
Future Works (1)
• Smarter Situations– Problem
• Crowd flow planning
– Solution• Situation Graph
– Represents aggregative relation between situations– Makes crowd follow a scenario– Provides interactive control– Controls the number of agents in a situation.
Situation Graph
Ticket booth(start)
Ticket booth(start)
Gather andTalk
Restroom
Movie room(end)
100
50
10
70
• Example
Future works (2)
• Hierarchical Situations– Problem
• Need to organize situations efficiently
– Solution• Hierarchical situations
– Organizes situations in a hierarchical way» e.g.) parent (queue), child (Vertical queue, Horizontal
queue)
Future Works (3)
• Hierarchical Environments– Problem
• Not easy to make a massive environment
– Solution• Hierarchical environment
Town
Theater
Lobby
Bench
Once we make a theater environment, we can copy and paste it to wherever we want.
Future Works (4)
• Adjustment of Discrete Action Choices
– Problem• Failed in satisfying constraints because of
shortage of discrete choices
– Solution• Provides a way to adjust actions to satisfy
constraints– e.g.) If an agent has a target position, we can adjust the
action choices to make agent move to the exact spot.
Future Works in Timeline
AdjustmentOf
Action Choices
HierarchicalSituation
HierarchicalEnvironment
SituationGraph
Jun/04
Sep/04
Dec/04
Mar/05