TeleSupervised Multi Agent Systems

1
Tele-Supervised Multi Agent Systems

• Kunal Sinha
• CSL
• 02/03/04

2
Structure of this talk

• Introduction to Multi Agent/Robot Systems
• Classification of Multi Agent/Robot Systems
• Teleoperation in Multi Agent Systems
• Proposed Scheme Tele-Supervised Multi Agent
  Systems
• Avenues for Implementation
• Future Work

3
Introduction

• Use of multiple robots working in coordination to
  execute different types of tasks can bring
  several advantages over a single robot solution
  suchas simplicity in robot design, better
  performance, increased fault tolerance and
  spatially distributed sensing and actuation
• Multi-robot teams can accomplish the same tasks
  using simpler and less expensive robots as
  compared to single robots
• They are more flexible and can be reconfigured
  and adapted to perform many tasks

4
Classification

• Size of the Collective
• SIZE-ALONE, SIZE-PAIR, SIZE-LIM, SIZE-INF
• Communication Range
• COM-NONE, COM-NEAR, COM-INF
• Communication Topology
• TOP-BROAD, TOP-ADD, TOP-TREE, TOP-GRAPH
• Communication Bandwidth
• BAND-INF, BAND-MOTION, BAND-LOW, BAND-ZERO
• Collective Reconfigurability
• ARR-STATIC, ARR-COMM, ARR-DYN
• Processing Ability of each collective unit
• PROC-SUM, PROC-FSA, PROC-PDA, PROC-TMA
• Collective Composition
• CMP-IDENT, CMP-HOM, CMP-HET

5
Classification

• Leader Follower Scheme
• Virtual Structure
• Behavioral Approach

6
Leader-Follower

• Leader has pre-defined trajectory
• Followers use its position and orientation as a
  reference to compute their desired position and
  orientation.

7
Leader-Follower

• Simplicity in the control structure
• The whole formation fails if Leader breaks down
• No feedback information
• Robot population not necessarily homogenous
• Applications
• Spacecraft and aircraft formations
• Cooperative manipulations
• Collaborative mapping
• Exploration

8
Virtual Structure

• Also known as Virtual Rigid Body or Virtual
  Leader
• Assumption All robots in the formation treated
  as one rigid object
• All robots maintain a fixed geometric position
  relative to others

9
Virtual Structure

• No distinguished Leader
• Formation and Motion control is centralized
• System not fault tolerant
• Demands continuous communication between the
  robots and the synchronization system
• Simplicity in task definition

• Homogenous robot population
• Necessity of use of external centralized
  coordination system
• Applications
• Spacecraft formations in free space
• Pushing or moving of objects by two or more
  mobile robots
• Robotic arm formations
• Satellite constellations

10
Behavioral Approach

• Simplification of solutions that exist in nature
• Formation behaviors in nature (flocking,
  schooling etc.) benefit the animals that use them
  (e.g. by minimizing individual encounters with
  predators)
• By grouping, animals also combine their sensors
  to maximize the chance of detecting predators or
  more efficiently forage for food.

• These behaviors emerge as a combination of a
  desire to stay in the group and yet
  simultaneously keep a separation distance from
  other members.
• Control function derived by averaging several
  atomic behaviors or specifying explicit
  dependencies
• if some condition
• then some action

11
Behavioral Approach

• Usually includes formation feedback
• Distributed and Scalable Control
• Difficult to analyze mathematically
• Applications
• Aircraft Flying
• Enclosing/tracing intruders
• Moving large number of small objects
• Collaborative mapping
• Exploration

12
Teleoperation in Multi Agent Systems

• Many purely reactive systems are myopic in their
  approach they sacrifice global knowledge for
  rapid local interaction.
• In essence, the teleoperator should be concerned
  with global social strategies for task
  completion, and should be far less involved with
  the specific behavioral tactics used by any
  individual agent.

• Reduction in the teleoperators cognitive and
  perceptual load by allowing the individual agents
  deal with their own local control concerns
• The teleoperator acts/interferes only as needed
  based upon observable progress towards task
  completion.

13
Teleoperation in Multi Agent Systems

• Single agent Teleautonomous Control
• Teleoperator as a schema
• Teleoperator as a supervisor
• Multi-agent Teleautonomous Control
• Tasks
• Foraging
• Grazing
• Herding

14
Teleoperation in Multi Agent Systems

• Foraging
• Wisely used Teleoperation can significantly lower
  the number of steps required to complete the task
  by greatly reducing the time spent in the wander
  state
• However, once the robots can sense an attractor,
  the teleoperator should stop giving instructions
  (unless they are needed to deal with a
  particularly troublesome set of obstacles).

• In general, the robots perform more efficiently
  by themselves than when under the teleoperators
  control if the agents already have an attractor
  in sight.
• Results Use of teleoperation as a guidance
  behavior resulted in 67 saving in terms of
  average number of time steps required for task
  completion

15
Teleoperation in Multi Agent Systems

• Foraging
• (a) Without Teleoperation
  (b) With Teleoperation

16
Teleoperation in Multi Agent Systems

• Grazing
• Robots performed poorly when large amount of
  teleoperation was involved
• Teleoperation only proved useful when the robots
  had difficulty in locating and ungrazed portion
  of the floor
• When used solely to help the robots find ungrazed
  floor area when they were not already cleaning,
  Teleoperation resulted in only 4 improvement in
  average task completion time performance

• Grazing Task

17
Teleoperation in Multi Agent Systems

• Herding

18
Tele-supervised Multi Agent Systems

• Multiple goals secured with optimal or near
  optimal allocation of agents.
• High level planning and prioritization of targets
  and decision making capabilities lying primarily
  with the Tele-assistant.
• The Tele-assistant concerned with the high global
  social strategies for task completion, and is far
  less involved with the specific behavioral
  tactics used by any specific agent in the group.
• No compromise on Global Knowledge in exchange for
  local interaction.
• Interference by Tele-assistant if and when
  desired to smoothen operations or resolve
  conflicts.
• Redistribution of bots into different groups,
  through Tele-assistant intervention, if a new
  target is identified during the securing process.

• Multi-tier control architecture with the
  complexity decreasing as one goes down from the
  Tele-assistant Level to the Individual Bots
  Level.
• May be extended to Tele-assisted driving of the
  Coordinating Slave(s) and Lead-Follower strategy.
• Capabilities for a growing formation based on
  leader selection and target prioritization.
• Complex multiple tasks to be accomplished with
  simple and inexpensive Robots.
• Flexibility in terms of configuration and
  adaptation to perform varied tasks.
• Flexibility in terms of Leadership assignment
  within a sub-group.

19
Tele-supervised Multi Agent Systems

• Throw 20 agents randomly
• Target identified
• Exits prioritized
• Choose Leaders
• Leaders Poll for Followers in the order of Target
  Priority
• Groups reassemble
• Groups move to secure respective exits
• Avoiding obstacles and other bots
• Following the leader to the goal
• Closing in with each other once there to secure
  exit

• New exit pops up
• Define priority for new exit
• Choose a Leader
• Since no independent bots
• Either every group donates some bots to the new
  group keeping in mind task priorities
• Or Tele-Supervisor chooses bots
• New group secures new exit

20
Tele-supervised Multi Agent Systems
21
Tele-supervised Multi Agent Systems
22
Tele-supervised Multi Agent Systems
23
Tele-supervised Multi Agent Systems
24
Tele-supervised Multi Agent Systems
25
Immediate Future Work

• Formalization of Control, Priority Allocation,
  Tele-assistance etc.
• Simulation platform?
• Matlab
• C
• MissionLabs
• Teambots
• Swarm

26
Thanks !

• Question/Comments/Concerns ??