PRP Presentation: A Bioinspired Group Evasion Algorithm

1
PRP PresentationA Bio-inspired Group Evasion
Algorithm

• Sang Woo Lee

2
Contents

• Problem definition
• Motivation
• Background
• Contributions
• Preliminary results
• Limitations
• Future works

3
Problem Definition

• Goal
• Simulate plausible crowd evasion behavior
• Crowd simulation
• Group behavior
• Crowd evasion behavior

4
Crowd

• A large number of people
• Frequently observed in modern society
• Train/Subway stations/Shopping mall
• Pedestrians in urban environment
• Mobs/Protestors/Marches

5
Human crowd simulation

• Simulates the movements/actions of a large number
  of people
• An important research topic in numerous fields
• Sociology, psychology
• Robotics, computer graphics
• Politics, environmental engineering,
  transportation

6
Human motion modeling

• Single human motion modeling
• Human locomotion based on bio-mechanical
  experiments
• Estimations of an articulated human body model
• Focused on detailed, low-level motions
• Crowd modeling
• People act and think differently in a crowd
• Form a social network via social influence,
  interaction, communication between group members
• Focused on high-level, coordinated group motions

7
Crowd modeling

• Particle Treuille06, Reynolds06
• Based on continuum perspective for the crowd
• Crowd motions as per-particle energy minimization
• Pros easy to scale up, global motion planning
• Cons motions of crowd are similar to fluid
  movement
• Multi-Agents
• Considers crowd as a large number of agents
• Pros control each agents movement
• Cons hard to scale up, hard to control group
  motion

8
Group behavior

• Individuals interact within group, and act in a
  coordinated way
• Different from individual behaviors
• Examples
• Flocking behaviorReynold87, Amato04
• Herd behavior
• Act in same way without planned direction
• Shepherding behaviorAmato04
• Panic behavior Helbing00
• Evasion behavior
• Mob behavior

9
Previous work

• Flocking behavior Reynold87
• Collision avoidance
• Velocity matching
• Match velocity with nearby individuals
• Flock centering
• Stay close to nearby individuals

10
Previous work

• Shepherding behavior Amato04
• One kind of flocking behaviors
• Single shepherd case

11
Previous work

• Panic behavior Helbing00
• Evacuation situation
• A force based model
• Based on social forces
• Calculated repulsive and frictional forces
  between agents

12
Crowd evasion behavior

• The group behavior that occurs when individuals
  try to escape dangerous situations
• Dangers
• Life-threatening
• Fire, tsunami, beasts, and disasters
• Violence
• Vehicles, terrorists, bombing
• Risk
• Arrest

13
Crowd evasion behavior

• Examples
• Evacuation from buildings
• Terrorist attacks
• Mobs/Protestors chased by the police
• Vehicles running into the crowd
• Beasts escaped from zoo attacking the people

14
Crowd evasion behavior
15
Difficulties of human crowd simulation

• No well-known scientific model exists
• Complexity of human group modeling
• Interactions, communication, and social influence
• More difficult than single human modeling
• Performing experiments is very difficult
• Complexity of dealing with a large number of
  agents
• Computationally intensive

16
Difficulties of human evasion behavior

• Less relevant evaluation data
• Due to the dangerous situations
• Only fire evacuation drill/survey data are
  available
• More difficult to model a human individual
• Direct experiments in dangerous situation are not
  available

17
Motivation

• Many crowd evasion patterns exist in modern
  societies
• Little research for crowd evasion has been done
• Only in building evacuation cases

18
Bio-inspired algorithms

• Bio-inspired algorithms
• Algorithms inspired by natural biological
  phenomena or model
• Algorithms
• Optimization, networking, genetic algorithm, game
  theories
• Robotics
• Terrain covering Svennebring03, Schwager08
• Redistribution robots Halasz07
• Collision avoidance using optic flow Barrows03

19
Group evasion model

• Well-researched evasion patterns in biology
• Predator-prey interaction research
• One of the most important research areas for
  biologists
• Patterns of predator evasion
• Predator evasion
• One of the most primitive danger evasions
• Even humans might not be exceptions

20
Group evasion model

• Imminent dangers allow no time to think
• In a high level perspective, a biological model
  should match with human evasion behavior
• In emergency situations, individuals start
  pushing and increase physical interaction
  Helbing00

21
Group evasion model

• Biological evasion model
• Good predator-prey model for fish school
• Well-observed evasion patterns
• Can simulate observed fish school evasion
  patterns
• Picture from Parrish02

22
My evasion model

• Agent-based model
• Based on two fish evasion models
• Mainly based on Inada02
• Able to simulate all the observed fish school
  evasion patterns except for the ball pattern
• Simple individual model with proper variables
• Limitations
• No obstacle in the space
• No speed increasing in urgent situations
• No angular speed constraints

23
My evasion model

• Integrate Zheng05 model
• Speed increases in urgent situations
• Angular speed constraints
• Urgent reactive field
• Use RVO as collision avoidance algorithm
• Added goal-driven movement
• Enable a group to move for certain goals

24
Reciprocal Velocity Object

• Local collision avoidance algorithmBerg08
• Change original velocities of agents to avoid
  collision

25
Biological evasion model

• Inada02s model
• Based on five motion rules
• Only determines moving direction
• Schooling rule Similar to Reynold87s
  Flocking
• Approach
• Parallel orientation
• Repulsion
• Substituted by RVO

26
Biological evasion model

• Inada02s model
• Rr Repulsive-orientation field
• RrRp Parallel-orientation field
• RpRa Attractive-orientation field
• w Visual angle

27
Biological evasion model

• Inada02s model
• Obedience level c
• The tendency to schooling motion rather than
  selfish evasion motion
• Change the frequency of evasion patterns
• Can simulate human individuals tendency between
  group decisions and selfish evasions
• Speed
• Observed by real fish school
• Gamma distribution

28
Biological evasion model

• Other motion rules
• Evasion
• Search
• Only when there is no other agent in reactive
  fields
• Only consider limited number of agents with front
  priority
• Move direction
• Vector summation of motion vectors from motion
  rules
• Add Gaussian perturbation

29
Biological evasion model

• Zheng05s model
• 1 Capture area
• 2 Urgent reaction area
• Evasion moving with urgency
• The moving speed of fish is faster than its usual
  speed
• 3 Caution area
• Evasion with no urgency
• 4 Patience area
• 5 Invisible area

30
Biological evasion model

• Zheng05s model
• Angular speed limitation
• Energy expenditure effect
• Not considered in my evasion model

31
My predator model

• Inada02s model
• Chasing limited number of prey
• Direction determined by vector summation
• Zheng05s model
• Chasing one target fish
• Add a sight occlusion
• My model
• Based on Vidal02
• Probabilistic approach with Markov random model
• Extensive model for future work

32
Contributions

• Evasion model integrated with two biological
  papers
• Novel approach in graphics and robotics
• Add goal-driven movement
• Extensive predator model
• Integrate my model to RVO framework
• Add two dimensional view frustum
• Testing with preliminary simulator
• Proper values for human individuals

33
Preliminary Results

• Using only a bio-inspired model
• Observe patterns with respect to changes with
  control variables
• What is the suitable values of human individuals?
  
• Visual angle
• Dynamic constraints
• Speed
• Angular speed
• Urgent speed factor
• Examine the possibility of applying a fish model
  to a human evasion model

34
Preliminary Results
35
Preliminary Results

• Exhibit proper flocking behavior
• Exhibit various evasion patterns
• Herd, split, hourglass, vacuole, flash expansion,
  flash turn, fountain effect
• Works for chasing predator

36
Evasion Patterns of fish school
37
Evasion Patterns of fish school
38
Evasion Patterns of fish school
39
Preliminary Results

• Other control variables
• Urgent area
• Motion is more natural with an urgent area
• Schooling behavior become weaker
• Predator pattern
• Two patterns in the biological papers are tested
• Inada02s model is more effective to represent
  evasion pattern
• The predators speed factor
• Evasion behavior is not presented properly if the
  predator is too fast

40
Preliminary Results

• Other control variables
• Parallel-orientation field
• Determine mean distance of the group members
• Goal-obedience level
• Low group steering is slow
• High jamming occurs

41
Preliminary Results

• Some result videos

42
Applications

• Computer Animation
• Virtual reality
• Military training center
• Disaster and emergency training
• Movie
• Video game

43
Limitations

• Biological model only works for imminent danger
• No communication, cooperation, complex behavior
  between agents
• Hard to use for complex scenarios
• Modeling evasion for distant dangers needs
  sociological/psychological factors

44
Future work(Ongoing work)

• Integrating sociological factors
• Find and design a proper sociological model for
  human evasion behavior
• Generalize a sociological crowd behavior model
• For other group behaviors

45
Future work(Ongoing work)

• Sociological crowd modeling
• Emergent norm theory Aguirre98
• Crowd are rational and norm-governed
• Not all predicable, but not random
• Social Impact Theory Latane96
• Social influence model at the individual level
• Ones behavior changes as a function of the
  strength, immediacy, number of sources

46
Thanks to

• Adviser
• Dinesh Manocha
• PRP committee
• Diane Pozefsky
• Kye S. Hedlund
• Anselmo A. Lastra

47
Reference

• Reynold87
• C. W. Reynolds. Flocks, herds and schools A
  distributed behavioral model. SIGGRAPH Comput.
  Graph., 21(4)25-34, 1987
• Amato04
• Shepherding Behaviors, Jyh-Ming Lien, O. Burchan
  Bayazit, Ross T. Sowell, Samuel Rodriguez, Nancy
  M. Amato, In Proc. IEEE Int. Conf. Robot. Autom.
  (ICRA), pp. 4159-4164, New Orleans, Apr 2004.
• Helbing00
• D. Helbing, I. Farkas, and T. Vicsek. Simulating
  dynamical features of escape panic. Nature,
  407487-490, Sep 2000.

48
Reference

• Coverage
• J. Svennebring and S. Koenig, Trail-laying
  robots for robust terrain coverage,, Proc. of
  IEEE International Conference on Robotics and
  Automation 2003, Volume 1, On page(s) 75- 82
  vol.1
• M. Schwager, F. Bullo, D. Skelly, and D. Rud, A
  Ladybug Exploration Strategy for Distributed
  Adaptive Coverage Control, Robotics and
  Automation, 2008, ICRA 2008
• Redistribution
• A. Halasz, M. Ani Hsieh, S. Berman, V. Kumar.
  Dynamic Redistribution of a Swarm of Robots Among
  Multiple Sites, 2007 IEEE/RSJ International
  Conference on Intelligent Robots and Systems.
• Collision avoidance
• Barrows, G.L., Chahl, J.S. and Srinivasan, M.V.,
  Biomimetic visual sensing and flight control.
  Aeronautical Journal. v107 i1069. 159-168

49
Reference

• Crowd Simulation
• D. Halmann, and S. Raupp Musse, Crowd Simulation,
  London, Springer, 2007
• Treuille, A. Cooper, S. Popovic, Z. Continuum
  Crowds. ACM Transactions on Graphics 25(3)
  (SIGGRAPH 2006)
• C.W. Reynolds, Big fast crowds on ps3. In
  sand-box 06 Proceedings of the 2006 ACM
  SIGGRAPH symposium on Videogames (2006),
  ACMPress, pp. 113121
• Biology
• M.Zheng, Y. Kashimori, O. Hoshino, K. Fujitai,
  and T. Kambara. Behavior pattern (innate action)
  of individuals in sh schools generating efficient
  collective evasion from predation. Journal of
  Theoretical Biology, 235153167, 21 July 2005.
• Y. Inada and K. Kawachi. Order and flexibility in
  the motion of sh schools. Journal of Theoretical
  Biology, 214371-387, 7 February 2002.
• J. K. Parrish, S.V. Viscido, and D. Grunbaum.
  Self-Organized Fish Schools An Examination of
  Emergent Properties, Biol. Bull. 202 296305.
  (June 2002)

50
Reference

• Robotics
• R. Vidal, O. Shakernia, H. J. Kim, D. Shim, and
  S. Sastry, "Probabilistic pursuit-evasion games
  Theory, implementation and experimental
  evaluation", IEEE Transactions on Robotics and
  Automation, Oct 2002.
• Sociology
• B. Latane and M. J. Bourgeois. Experimental
  evidence for dynamic social impact The emergence
  of subcultures in electronic groups. The Journal
  of Communication, 4635-47, 1996.
• B. E. Aguirre, D. Wenger, and G. Vigo. A test of
  the emergent norm theory of collective behavior.
  Sociological Forum, 13(2)301-320, 1998.