Dynamic vehicle routing using Ant Based Control

1
Dynamic vehicle routingusing Ant Based Control

• Ronald Kroon
• Leon Rothkrantz
• Delft University of Technology
• October 2, 2002
• Delft

2
Contents

• Introduction
• Theory
• Ant Based Control
• Simulation environment and Routing system
• Experiment and results
• Conclusions and recommendations

3
Introduction (1)

• Dynamic vehicle routing
• using Ant Based Control

• Routing cars through a city
• Using dynamic data
• Using an Ant Based Control algorithm

4
Introduction (2)
Goals

• Design and implement a prototype of dynamic
  Routing system using Ant Based Control
• Design and implement a simulation environment for
  traffic
• Test Routing system

5
Introduction (3)
Possible applications

• Navigate a driver through a city
• Find the closest parking lot
• Divert from congestions

6
(No Transcript)
7
(No Transcript)
8
Schematic overview of the PITA components
9
3D Model of dynamic traffic data
10
Theory (1)

• Natural ants find the shortest route

11
Theory (2)

• Choosing randomly

12
Theory (3)

• Laying pheromone

13
Theory (4)

• Biased choosing

14
Theory (5)
3 reasons for choosing the shortest path

• Earlier pheromone (trail completed earlier)
• More pheromone (higher ant density)
• Younger pheromone (less diffusion)

15
Ant Based Control (1)
Application of ant behaviourin network management

• Mobile agents
• Probability tables
• Different pheromone for every destination

16
Ant Based Control (2)

• Probability table

(Node 2) Next 1 3 5
Destination
1 0.90 0.02 0.08
3 0.03 0.90 0.07
4 0.44 0.19 0.37
5 0.08 0.05 0.87

17
Ant Based Control (3)
Forward agents

• Generated regularly from every node with random
  destination
• Choose route according to a probability
• Probability represents strength of pheromone
  trail
• Collect travel times and delays

18
Ant Based Control (4)
Backward agents

• Move back from destination to source
• Use reverse path of forward agent
• Update the probabilities for going to this
  destination

19
Ant Based Control (5)
Updating probabilities

• Probability for choosing the node the forward
  agent chose is incremented
• Depends on
• Sum of collected travel times
• Delay on this path
• Update formula ?p A / t B
• Probabilities for choosing other nodes are
  slightly decremented

20
Simulation environment and Routing system (1)

• Architecture

21
Simulation environment and Routing system (2)

• Communication flow

22
Routing system (1)
23
Routing system (2)

• Timetable

1 2 4 5
1 - 12 15 -
2 11 - - 18
4 14 - - 13
5 - 18 14 -

24
Routing system (3)

• Update information

t1
25
(No Transcript)
26
Simulation environment (1)

• Map of Beverwijk

27
Simulation environment (2)

• Map representation for simulation

28
Simulation environment (3)
Simulation with driving vehicles
29
Simulation environment (4)
Features

• Traffic lights
• Roundabouts
• One-way traffic
• Number of lanes
• High / low priority roads

• Precedence rules
• Speed variation per road
• Traffic distribution
• Road disabling

30
Experiment
31
Results
In this test case (no realistic environment)

• 32 profit for all vehicles, when some of them
  are guided by the Routing system
• 19 extra profit for vehicles using the Routing
  system

32
Conclusions

• Successful creation of Routing system and
  simulation environment
• Test results
• Routing system is effective
• Smart vehicles take shorter routes
• Other vehicles also benefit from better
  distribution of traffic
• Routing system adapts to new situations
• 15 sec 2 min

33
Recommendations

• Let vehicle speed depend on saturation of the
  road
• Update probabilities using earlier found routes
  compared to new route
• Use the same pheromone for all parkings near a
  city center

34
Start demo
Demo