Adrian Treuille, Seth Cooper, Zoran Popović2006
Walter Kerrebijn045837607-06-2011
Introduction
Crowd Motion:• large groups • common goals• collision avoidance• real-time• natural
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:• “Real-time motion synthesis model for large crowds without agent-based dynamics”
Introduction
Motion:• per-particle energy minimization• dynamic potential and velocity fields• merge of local and global path-planning
Related Work
Methods in Game Development:• Grid-based• Navigation Meshes• Waypoint GraphCombined with reactive steering approach
Continuum Crowd Approach
Each person in a crowd:1. is trying to reach a goal2. moves at the maximum speed possible3. tries to avoid discomfortable areas4. picks the path minimizing the
weighted sum of 1. and 2. and 3.
Hypotheses
Continuum Crowd Approach
Static goals can for example be:• go to specific address• go to ‘west side’ of town
Dynamic goals can for example be:• follow specific person• find (non-)empty theater seat• explore unseen parts of environment
Hypotheses
Continuum Crowd ApproachHypotheses
Maximum speed depends on:• environment• other people
Continuum Crowd ApproachHypotheses
Avoiding discomfort fields encourages people to take certain paths
Continuum Crowd ApproachHypotheses
Continuum Crowd ApproachHypotheses
Continuum Crowd ApproachOptimal Path
Continuum Crowd ApproachOptimal Path
Continuum Crowd ApproachOptimal Path
Calculating a potential field may be done simultaneously for a group of characters
Continuum Crowd ApproachSpeed
Speed is depending on:1. crowd density2. terrain
Continuum Crowd ApproachSpeed
Crowd Density Field
Average Velocity Field
Continuum Crowd ApproachSpeed
For areas of low crowd density, the speed is depending on the terrain
Continuum Crowd ApproachSpeed
For areas of high crowd density, the speed is depending on the crowd
Continuum Crowd ApproachSpeed
For areas of medium crowd density, the speed is depending on both the terrain and the crowd
Continuum Crowd ApproachPrediction
Some predictive measures are necessary to reduce unnatural behavior:
• Predictive Discomfort- adds future density to discomfort field- should deal with perpendicular crossing
• Expected Periodic Field Changes- calculates expected speed - should deal with situations like traffic lights and doors
Implementation
The algorithm used is as follows:
Algorithm
Implementation
The algorithm used is as follows:
Algorithm
ImplementationDensity Conversion
ImplementationDynamic Field Construction
Choosing least cost neighbor:
Finite difference approximation:
Experiment
2D and 3D setups3.4 GHz Nvidia Quadro FX 3400
[Movie]
Results/Conclusion
• Simulation steps took between 2 and 5 fps (?)
• Human animations were too simple
• Vortices and lanes emerged
• Agent interaction was possible
• Minimum Distance Enforcement was necessary
Assessment
The idea to merge local and global path planning is nice, but is it really better?
• weird behavior at traffic lights• collisions still happen• discomfort does not behave ‘natural’ enough• individual control is lost• there is no apparent group identity/cohesion
Does this method more closely resemble human psychology and path planning?
• global and local goals• wandering
Assessment
It is not clear how the grid size is chosen, or how its choice influences the system
It is not clear why there is a ‘hard cut’ between low, medium and high crowd density speed calculations
There is no real mention of goal selection
Assessment
The experimental setup was not merged with an agents approach (as mentioned in the paper), only compared against it, so there is no way to see agent interactions with continuum crowds
FPS is not a measure of time, so how to evaluate these experiments?
Assessment
The results did not include tables, graphs or any other data visualization