Hengchin Yeh, Sean Curtis, Sachin Patil, Jur van den Berg,
Dinesh Manocha, Ming LinUniversity of North Carolina at Chapel Hill
ACM 2008
Walter Kerrebijn045837621-06-2011
Introduction
Increase of agent-based methods to model virtual crowds:
• off-line (movies)• real-time (games, virtual environments)
Introduction
Agent-based approach pros:• independent decisions• different simulation parameters
Agent-based approach contras:• emergent realism from behavioral rules hard to ensure• computationally expensive• distinction between global and local path-planning
Introduction
Proposal:• Use composite agents to model different emergent behaviors:
- embody intangible factors (social, psychological)- use pre-existing collision avoidance
Related Work
• Rule-based systems• Social Forces models• Continuum Crowd theoryClaim: All these can be combined with Composite Agents approach
Composite Agents
General multi-agent system (SIMULATOR):• environment ΦEnv
• set of Agents = {A1,A2,…,An}• with states φi
• external state εi• position pi• velocity vi• geometric representation Gi
• internal state ιi• goal position, memory, mental state
Definitions
Composite Agents
General multi-agent system (SIMULATOR):• Algorithm for each agent:
• GatherNeighbors()• field of view, nearest-k neighbors• ENbr = {εk | Ak є GatherNeighbors(Ai)}
• Update()• φi ← Update(φi,ENbr,ΦEnv)
Definitions
Composite AgentsDefinitions
Composite Agent:• Basic Agent
• standard agent Ai from SIMULATOR• contains a set of Proxy Agents Pi,j
• Proxy Agent• “hands extended from the basic agent […], encouraging [other agents] to step away to avoid collision”
Composite AgentsDefinitions
Proxy Agent Pi,j• εi,j• ιi,j• acces to ιi
Composite AgentsDefinitions
Composite AgentsTypes
Different kinds of intangible factors:• Aggression• Social Priority• Authority• Protection and Guidance
Composite AgentsTypes
Aggression:• Urgency
• modeled as property Urgency• Expression of that urgency
• modeled by adding aggression proxy Pi,1
Composite AgentsTypes
Urgency:• constant
• dynamic (velocity-based, distance-based)
Composite AgentsTypes
Example Urgency
Composite AgentsTypes
Social Priority:• Priority
• modeled as property Priority• Expression of that priority
• modeled by adding priority proxy Pi,1
Composite AgentsTypes
Example Social Priority
Composite AgentsTypes
Authority:• Trailblazer
• modeled as property Trail Identifier• Expression of that trailblazer
• modeled by adding trail proxies Pi,1,Pi,2,…,Pi,m
Composite AgentsTypes
Example Authority
Composite AgentsTypes
Protection and Guidance:• Mother M and Child K
• M maintains information about K• M provides protection and guidance for K
• Expression of M’s behavior• modeled by adding a protection or guidance proxie Pi,1
Composite AgentsTypes
Protection:
Guidance:
Composite AgentsTypes
Example Protection and Guidance
Implementation
Implementation
Implementation
Proxy Updates• information contained in proxy
Dynamic StatesConditional Neighbors
• proxies not in neighbor set of parent agent, trail proxies not in neighbor sets of group members
Visualization• 2D and 3D
Experiment
Office Evacuation, Subway Station, Embassy
[Movie]
Results
Results
Results
Conclusion
• Composite agents can be succesfully used to model emergent crowd behaviors• This yields little computational overhead
Assessment
• (Almost) good paper length, but lacking information almost everywhere
• Experiments barely compare between methods or even sufficiently in the same method
• Ending seems too short, incomplete, or superficial
• Conclusion is not epic, and maybe too bold
Assessment
• The ‘math’ section seems misplaced and arbitrary, also too compact to really check its use and correctness
• Almost nothing is mentioned about goal selection, map creation, or the selection of locations of proxy agents
• Accompanying website (http://gamma.cs.unc.edu/CompAgent/) has very little information
Assessment
• The notion of ‘groups’ is not really explored
• ‘Any geometrical shape’ is not explained
• ‘Future work’ should be current work