MultiAgent Systems

1
Multi-Agent Systems

• Cooperation and Collaboration

2
Motivation

• Designing Teams is hard
• More agents can interfere with eachother
• Team performance is hard to monitor
• Is communication needed ?
• Individual autonomy or centralised control

3
Heterogeneity

• Degree of similarity of individuals
• Homogenous teams
• identical individuals, but possibly different
  roles and behaviours
• easy to maintain and program
• easy to replace failed robot

4
Heterogeneous Teams

• Team leader homogeneous team
• leader coordinates teams
• more powerful processing, software and hardware
• Single point of failure
• e.g.
• Ground team controlled by helicopter
• Marsupial robots, team carried by leader

5
USF Marsupial Robot
6
Social Entropy

• Measure of how different the team members are
• Het(R) -?ci1 pilog2(pi)
• where R is the team
• c is the number of different types of robots
• and pi is the proportion of robots of each type

7
Control

• Centralised vs distributed
• Centralised
• top down, individuals follow leaders commands
• Distributed
• individuals decide their own behaviour. System
  behaviour emerges

8
Cooperation

• Active Cooperation
• individuals acknowledging and working together
• Non-active Cooperation
• treating other individuals as obstacles can lead
  to cooperative behaviour
• Physical Cooperation
• robots actually interact physically, carry
  eachother, puch each other etc, to achieve goals

9
Goals

• Single team goals
• All robots explicitly aim to achieve a single
  goal
• Individual goals
• Each robot has a single goal that may help
  achieve team goal
• Emergent cooperation
• not same as a single team goal

10
Emergent Social Behaviour

• Ignorant coexistence
• everything is an obstacle
• Informed coexistence
• give way to robots
• Intelligent coexistence
• head in the groups direction, flocking behaviour
• Maja Matarics Nerd Herd

11
Robot Motivation

• Robot impatience
• Other robots performance poor, impatience
  increases
• Robot acquiescence
• Own performance poor acquiescence increases.
• Lynne Parkers ALLIANCE architecture
• extension of subsumption architecture

12
Case Study- Robot Soccer