Wouter G. van Toll

1
Realistic Crowd Simulationwith Density-Based
Path Planning

• Wouter G. van Toll
• Atlas F. Cook IV
• Roland Geraerts

ICT.OPEN / ASCI October 22nd, 2012
2
Introduction

• Path planning in virtual environments (e.g.
  games)
• Global planning find a main route
• Local planning variety, collision avoidance
• Many characters at once crowd simulation
• Problem in a crowd, short routes are not always
  good
• Collision avoidance cannot solve everything
• Other routes are unused

3
Overview

• We use crowd density information for global
  planning
• Stored in a navigation mesh (Explicit Corridor
  Map)
• Planning algorithm time-based A
• Periodic replanning
• Results
• Characters take detours around congested areas
• Crowd spreads over different routes
• Efficient tens of thousands of characters in
  real-time

4
Sneak preview

• Before After

5
Preliminaries

• Navigation Meshes / Crowd Density

6
Navigation meshes

• Characters need to find paths through an
  environment
• Navigation graph 1D edges
• Not flexible enough for crowds
• Navigation mesh 2D polygons
• Global path sequence of polygons
• Local planning during movement
• Common in modern games / simulations
• Crowd simulation has been solved!
• Assumptions in the navigation mesh / crowd
• General framework?

7
Explicit Corridor Map

• An exact and flexible navigation mesh
• Medial axis
• Closest-obstacle annotations
• A on the medial axis ? path
• corridor
• Shortest paths with clearance
• Collision avoidance
• Multi-layered environments
• Dynamic updates

8
Crowd density

• Fraction of a region Ri that is occupied by
  characters
• Often in persons per m²
• Often in levels and colors
• For us value ?i between 0 and 1
• (allows multiple character sizes)
• Practical studies
• When the density is high,people walk more slowly
• Density-speed function v(?) 0,1 ? 0, vmax
• Time-based path planning?

F
E
D
C
B
A
9
Method

• Density Map / Density-Based Path Planning

10
Density map

• ECM divides the environment into non-overlapping
  regions
• Each region maps to an ECM edge
• Each region stores its local density
• Updated in real-time
• Density of a region
• ? Expected walking speed within the region
• ? Expected traversal time of the edge
• ? Expected delay

11
Density-based path planning

• A search on the medial axis
• A character wants a fast path, not necessarily
  the shortest
• Each ECM edge ei has a density ?i
• tmin(ei) traversal time at speed v(0) vmax
• tdens(ei) traversal time at speed v(?i)
• tdelay(ei) tmin(ei) - tdens(ei)
• cost(ei) tmin(ei) w tdelay(ei)
• Controlling the sensitivity to delays
• w 0 shortest path
• w 1 fastest path (Höcker et al., 2010)
• w gt 1 more eager to take detours

12
Replanning

• Densities change all the time
• Characters should re-check their paths
• We cannot see crowds that are far away
• Density viewing distance D along the medial axis
• A if the path length gt D, assume ? 0
• ? Path has a visible and an invisible part
• Partial replanning re-use invisible parts
• Character has moved
• More points are visible
• At a mutually invisible point, A can stop
• ? Speed vs. knowledge

13
Results

• Realistic Crowds in Real-Time

14
Experimental results

• Varying the cost weight w
• w 0 ? congestions
• w 1, no replanning ? periodic effect
• w 1, replanning ? realistic crowd flow
• ? longer but faster paths
• w gt 1 ? larger detours, indecisive crowd
• ? Useful in other environments?
• Varying the viewing distance
• 1 - 2 ms in a large city
• Real-time periodic replanning
• Multi-threading steering 50K characters in 30
  ms/frame
• 70 ms/frame left for e.g. replanning

D 8
D 350 m
D 0 m
15
Closing Comments

• Conclusions / Future work

16
Conclusions

• The ECM navigation mesh serves as a density map
• Non-overlapping, exact, always defined
• Region densities map to edge speeds
• Updated in real-time
• Characters use density-based A
• Short paths with little expected delay
• Sensitivity to delay can be set
• (Partial) replanning
• Result More realistic crowds
• Characters spread over routes
• Characters avoid congestions
• Emergent global effects due to individual
  choices
• (Still) real-time performance for large crowds

17
Future work

• New questions
• Use flow information? (speed direction)
• Use actual visibility?
• Event-based vs. periodic replanning
• Plan with terrain information?

18
More information

• Poster
• Contact us
• W.G.vanToll_at_uu.nl
• R.J.Geraerts_at_uu.nl
• http//people.cs.uu.nl/roland/
• Questions?