ant_robots

1
ANT ROBOTS
Information Theoretic Self Organization of
Multiple Agents
2
Presentation Outline

• Multiple Agents An Introduction
• How to build an ant robot
• Self-Organization of Multiple Agents
• Collective Target Tracking
• Towards real-time implementation
• Conclusion

3
Multiple Agents

• What are Agents?
• Agents are man-made entities which can perform a
  particular task.
• Why Multiple Agents?
• Some tasks may be inherently too complex or
  impossible for a single agent to perform
• Several simple agents may be easier to
  design/cost efficient than a single complex
  agent.
• Flexibility and Robustness
• Comparable to pack hunters

4
Multiple Agents

• More generally called as swarms of agents.
• Swarm Intelligence Evolved from studies of
  insect societies
• Ex. Ant foraging activity is used to solve
  problems in big communication networks
• Complex collective behavior from the
  interactions of simple individuals
• Wide-range application ex. material
  transportation, planetary missions, oceanographic
  sampling

5
Multiple Agents

• Future of Multiple Agents Nanoswarms
• Swarm intelligence combined with Nanotechnology
• Also called as Molecular Robotics
• NSF has recently funded a research at USC
  (University of Southern California) to build
  nanoswarms for controlling water pollution.

"AI" to "AL"

6
Multiple Agents

• Self-Organization of Multiple Agents
• Centralized Vs Decentralized Approach
• Disadvantages of Leader Follower strategy
• Fragile
• Extensive communication
• May not be feasible
• Cost considerations

7
Multiple Agents

• Biological Inspiration
• Ants are fascinating social insects. They are
  only capable of short-range interactions, yet
  communities of ants are able to solve complex
  problems efficiently and reliably. Ants have
  therefore become a source of algorithmic ideas
  for distributed systems where a robot (or a
  computer) is the "individual" and a swarm of
  robots (or the network) plays the role of the
  "colony".

8
Multiple Agents

• Proposed technique
• Decentralized self-organization of Multiple
  Agents with minimal hardware using Information
  Theoretic Interactions
• Goals
• Distribute a group of robots uniformly inside a
  region
• Move the collective towards a target

9
How to build an ant-robot
10
How to build an ant-robot
11
How to build an ant-robot

• We can add any kind of ant head we like.
• Note The ant needs quite a lot of power to lift
  itself up and down as it walks. We need to use
  fresh batteries in our motor's power box. If you
  reverse the direction the motor spins (by using
  the switch on the power box), the ant will walk
  forwards and backwards.

12
Self-Organization Algorithm

• Goal 1 Spreading the robots uniformly over a
  region
• Spreading of agents Literature Survey
• Centralized control
• Nearest neighbor(s) repulsion
• Detecting distance from the farthest agent
• Assuming a central beacon
• Simplest region A circle is chosen as the first
  step

13
Self-Organization Algorithm

• Calculating the controlling force needed at the
  boundary of the circle is a complex optimization
  problem.
• Not equivalent to the packing of identical
  circles inside a circle problem

14
Self-Organization Algorithm

• The controlling force needed at the boundary of
  the circle is calculated empirically.

• Threshold

15
Self-Organization Algorithm

• The potential Vi for a particular robot is the
  sum of the received signal amplitudes Aik from
  all other robots.
• Vi is then compared to the threshold ? to
  specify the sign of the Information force (IF)
  given by

16
Self-Organization Algorithm
Simulation Video for spreading and obstacle
avoidance
17
Self-Organization Algorithm
Extending the algorithm to any region bounded by
piecewise-linear boundary Jordan Curve Theorem
18
Self-Organization Algorithm
Simulation Video for spreading over a star-shaped
region
19
Collective Target Tracking

• Goal 2 Move the collective towards a target
• Before detecting the ant food in the middle of
  ant robots.
• After detecting the ant food all ant robots
  gathered near it.

1
2
20
Collective Target Tracking

• Four situations for this case
• Situation 1 All the robots know the position of
  the target and their own positions. Then, its a
  trivial problem.
• Situation 2 Only one leader robot knows the
  position of the robot and it leads the group
  towards the target.
• Disadvantage Centralized Control
• Possible remedies Have many leaders, Rotate
  the leader
• Situation 3 The target issues a beacon signal.
  The group moves towards it.
• Situation 4 The robots does not know the
  position of the robot and decentralized approach
  is needed. Best assumption in military
  applications.

21
Collective Target Tracking

• Continuing with Situation 4
• Two base stations are used to transmit direction
  information to all robots by giving separate IF
  to all the robots.
• Each base station is assumed to have a simple
  radar.
• Radar decides whether the center of the circle
  is to the left or right of the target Line of
  Sight.
• The direction of the IF is rotated accordingly.
  IF from the base stations is same for all the
  robots.
• Total IF experienced by the robots is the sum of
  IFs between the robots and due to the base
  stations.

22
Collective Target Tracking
Rotation of IF from the base stations
23
Collective Target Tracking
Simulation Video for moving target tracking
24
Towards Real-time Implementation

• Various parameters used in the current
  simulation
• Number of robots 6
• Size of the robot 16cm ? 9cm
• Differential wheels
• Axle length 16cm Wheel radius 5cm
• Min speed 1 rad/s Max speed 10 rad/s
• Acceleration 10 rad/s2
• Radius of the circle 3m
• Time taken to spread 16 s

25
Towards Real-time Implementation
Simulation in Ant Robots
26
Conclusion

• Use of Information theoretic interactions to
  self-organization of multiple agents gives a good
  and simple solution for spreading of the robots
  and moving them towards a target.
• The algorithm can be suitably changed for the
  intended application.
• Collision Avoidance is naturally built-in during
  spreading and tracking.

27
Conclusion

• The decentralized Self-Organization Algorithm is
  cost-effective since
• Each robot needs to have only a simple
  transmitter and receiver.
• The circuit complexity of the receiver does not
  increase with increasing the number of robots.
• It is possible to make each robot not know its
  own absolute position as well as the position of
  the other robots.