Obstacles to Producing Realistic Urban Traffic in a Distributed Environment

1
Obstacles to Producing Realistic Urban Traffic in
a Distributed Environment

• By Dale Moyer
• Lockheed Martin Simulation, Training, and Support

2
Getting Cultured
(What is CultureSim?)

• Generates civilian populace on a large
  scale(1,000,000 entities)
• Provides sensor clutter
• Simple Creation
• Autonomous

3
Why is this hard?

• Quantity
• Initial target 10,000 vehicles (1999)
• Initial capability 18 vehicles per 200 Mhz
  computer
• Task excludes aggregate model
• Large sensor footprints
• Realistic behavior required
• High fidelity terrain with road network
• Close zoom 2d views, 3d views
• Individual behavior will be monitored to
  determine if enemy or civilian

4
Divide and Conquer

• Trim vehicle state and behaviors
• Still too much
• Divide vehicles between simulators
• Track movement by road
• Quick, finely divided sorting and storage
• Collision avoidance in linear time instead of N2
• Lanes do thinking instead of vehicles more
  scalable w. dense traffic

5
Divide and Conquer

• Network expenses
• Limited bandwidth
• CPU expense of processing entity state updates
• Geographic filtering
• DDM offers network filtering by location
• Static Grid
• Limited Granularity
• Wandering vehicles break

6
Solution 1 Land of the Blind

• Do not subscribe to entity state
• Message when vehicle enters road
• Unfiltered
• Small
• Infrequent
• Deterministic road travel
• Huge expense reduction, but less scalable
• Network traffic unfiltered
• Every CultureSim simulates every intersection

7
Problems

• Out-of-sync
• Uncertainty of location
• De-ordered lists
• Parking
• Cascading desynchronization
• Cant change behavior mid-road

8
Solution 2 Road Ownership

• Each sim watches a portion of the roads
• Geographically co-located roads
• Listen to state of vehicles in area
• Filter using DDM
• Send travel permissions to sims of vehicles in
  area
• Essentially, updating attributes of remote
  objects
• Extra network traffic per vehicle, but filtered
• Entity state
• Larger
• Changes more frequently
• Travel Permission
• Ultimately more scalable

9
Latency

• Apparent collisions
• Teleporting
• Cant always smooth
• Userconfusion
• Simulation accuracy

10
Road Based Dead Reckoning

• Describe movement in terms of road travel instead
  of world coordinates
• Prevent overruns by sending stopping distance in
  advance
• Fewer updates
• Free acceleration
• Turning and road entry/exit free
• No stop update
• Drawback reduced interoperability
• Complex, hard to implement
• Requires shared terrain topography

11
Road Based Dead Reckoning

• Road DR Info
• Current Target Road
• Current Distance Speed
• Stopping Distance
• Offroad entry/exit position
• Travel Permission Update
• Road distance to stop at
• Parking info, if exiting road

12
Additional Obstacles

• Terrain limitations
• Partial and/or broken road data
• Lack of traffic control data(lights, signs,
  lanes, directionality, etc)
• Bad scenario design
• Too many
• Nonsensical behavior mixes
• Poorly selected routes
• Traffic jam!

13
Solution Smarter Pathing

• Roads report average wait
• Plan based upon expected total travel time, not
  distance
• Replan mid-route if conditions change

14
Realistic Behavior Commuter Culture

• Create a schedule for entities
• Use building type information to produce
  realistic behavior
• Unique (home, work) and non-unique (restaurant,
  entertainment) events
• Actions can be based upon day of the week, time
  of day, probability, etc
• Produces collective traffic patterns, too
• Rush hour!
• Varied conditions without needing separate
  scenarios

15
Commuter Culture Profile
16
Future Improvements

• Ownership management
• Less network traffic
• Reduces impact of latency
• Isolate Road DR
• Make separate library for easy integration
• Output processed terrain file
• Still have correlation issues

17
Questions?