Difference between revisions of "Plan Rehan O'Grady"

From IridiaWiki
Jump to navigationJump to search
 
(32 intermediate revisions by the same user not shown)
Line 1: Line 1:
  +
==Meeting 15th June 2006==
==Approach==
 
   
  +
==='Swarm Control' - Theme for phd thesis===
#Design Experiment
 
#Get Input/Confirmation from Marco
 
#Build Controller
 
#Do Experiment
 
#Publish
 
   
  +
;Rationale
  +
* For the forseeable future the levels of autonomy in any human assisting (robotic) tool will be such that human direction at some level will still be required. e.g. sheep dog. Highly autonomous 'tool'. Still needs direction by human signals.
  +
* Giving commands to swarm robots particularly difficult. No simple mapping from human goals to robot internal representations.
  +
* Nice tie in with Swarmanoid - here it is the eyebot swarm giving commands to other swarms. (This is more complex, as the command givers are also distributed).
   
  +
==Long Term - Phd Theme==
 
  +
;Initial Research Areas
;Explore aspects of functional self assembly.
 
  +
* Basic control of swarm. Human (Eye-bots in swarmanoid?) can see the robotic swarm, an object to be retrieved and the target location. Need to direct the swarm to the object, then to the target location to which the object needs to be transported.
  +
* A command is broadcast to the swarm. The swarm must use distributed control mechanism to execute the command in an appropriate way. Example of a single broadcast command: '20 robots go over there and grab that object, 30 go and see what is over that hill. Rest of you stay where you are.'
  +
* Which robots receive commands? The whole swarm or just a selected few 'leader robots'? Considerations of expense (communication equipment) and efficiency. How do 'leader robots' direct the swarm?
  +
* Command modalities (longer term research) - console, speech, gesture, sound, light shining, clapping, remote controller (IR, sound - dog whistle).
  +
  +
  +
;Tie in with my work to date
  +
* Functional self-assembly is a form of swarm decision making. This ties into HSRI (human swarm-robot interaction) because an essential part of swarm control system design will be to determine to what level the swarm can / should behave autonomously.
  +
* All my work to date can be used as behavioural components that can be launched by the control mechanisms I will design over the Phd. e.g. Command is : 'Head to that light. If you meet an obstacle functionally self-assemble into groups of size 4'. OR 'Head to that light. If you meet an obstacle get into pairs and adaptively rotate over the hill'.
  +
  +
  +
===Shorter Term===
  +
  +
* Finish two sbot hill hill rotation experiment and write it up for a paper.
  +
* Pattern generation experiment with Anders - parameterised controller that can generate different patterns (chains, circle etc). Write up as a paper for a low key conference. I can use this simple controller as a starting point for investigating swarm control. Anders can use this controller as a starting point to investigate swarm fault detection.
  +
<br>
  +
  +
==Meeting 2nd December 2005==
  +
  +
 
===Long Term - Phd Thesis===
  +
;Theory
 
:Explore aspects of functional self assembly.
 
#Decision Process
 
#Decision Process
#*Is self-assembly necessary
+
#*Is self-assembly necessary?
 
#Structure
 
#Structure
#*What is appropriate size for connected groups
+
#*What is appropriate size for connected groups?
#*What is appropriate shape for connected groups
+
#*What is appropriate shape for connected groups?
 
#Behaviour
 
#Behaviour
 
#*What to do when connected (cooperative movement, transport, navigation, etc)?
 
#*What to do when connected (cooperative movement, transport, navigation, etc)?
Line 21: Line 44:
 
#*When should swarm disassemble?
 
#*When should swarm disassemble?
   
  +
;Practice
  +
#Parametrised group size selection mechanism (2,3,4,..)
  +
#*Initially in a non-adaptive way
  +
#*Maybe in an adaptive way
  +
#Parametrised Shape distinction mechanism
  +
#*Initially binary parameter - linear / blob (non-linear)
  +
#*Initially non-adaptive
  +
#*Maybe adaptive
  +
#*Maybe further parameterisable - control over what 'blob' looks like
  +
#Application of above two mechanisms to hill passing problem
  +
#*Comparison of adaptive linear approach and blob approach
  +
<br>
   
==Short Term (~End March 2006)==
+
===Short Term===
  +
====Timescale====
 
  +
Conference paper writtern by end March 2006
Goal: expand functional self-assembly experiment and publish new and old work together in a journal.
 
   
===Experiment===
+
====Goal====
  +
Conduct new experiment which builds on previous functional self-assembly experiment. Publish new results in conference (probably Ants2006). Combine new conference paper with ecal2005 paper and publish results in journal.
   
 
====Experiment====
Environment: Same as previous functional self assembly experiment.
 
   
  +
;Environment
;Goal
 
 
*Similar to previous ecal2005 functional self assembly experiment.
*''Large'' group of robots must navigate over hill towards target.
 
  +
*Except use three different hills to prove adaptive rotation mechanism. <code>| / \ </code>
*If they encounter hill, they aggregate into several two sbot swarmbots (using group size selection mechanism)
 
  +
*The two-sbot swarmbots separately navigate the hill into the target zone. (using adaptive swarm rotation to avoid parallel approach)
 
  +
;Task
 
*''Two'' robots must navigate over hill towards target.
  +
*When they encounter hill, they use adaptive rotation (adaptive to orientation of hill) to avoid toppling.
   
 
;Methods
 
;Methods
 
*Reuse existing functional self assembly controller. (Published work)
 
*Reuse existing functional self assembly controller. (Published work)
  +
*Develop adaptive (to hill orientation) control. (New work)
*Group size selection mechanism to form two sbot swarm bots. (Completely new work)
 
*Use adaptive swarm rotation mechanis to avoid parallel approach problem. (Previous unpublished work)
 
   
  +
;Trials
==Medium Term (~September 2006)==
 
  +
*2 robots: 20 trials on each hill type. = 60 total.
  +
*3 robots (maybe): some trials needed to show blob behaviour doesn't change with different hill orientations.
  +
<br>
   
  +
===Misc Ideas===
*Goal: Further experimentation to add body to Phd / publish.
 
  +
*Spend 1 or 2 weeks helping Lausanne disassemble / repair robots
  +
**I get to learn more about robotic hardware
  +
**Lausanne gets work done faster with extra pair of hands.
  +
**Iridia gets skill transfer on hardware disassembly / repair.

Latest revision as of 10:54, 16 June 2006

Meeting 15th June 2006

'Swarm Control' - Theme for phd thesis

Rationale
  • For the forseeable future the levels of autonomy in any human assisting (robotic) tool will be such that human direction at some level will still be required. e.g. sheep dog. Highly autonomous 'tool'. Still needs direction by human signals.
  • Giving commands to swarm robots particularly difficult. No simple mapping from human goals to robot internal representations.
  • Nice tie in with Swarmanoid - here it is the eyebot swarm giving commands to other swarms. (This is more complex, as the command givers are also distributed).


Initial Research Areas
  • Basic control of swarm. Human (Eye-bots in swarmanoid?) can see the robotic swarm, an object to be retrieved and the target location. Need to direct the swarm to the object, then to the target location to which the object needs to be transported.
  • A command is broadcast to the swarm. The swarm must use distributed control mechanism to execute the command in an appropriate way. Example of a single broadcast command: '20 robots go over there and grab that object, 30 go and see what is over that hill. Rest of you stay where you are.'
  • Which robots receive commands? The whole swarm or just a selected few 'leader robots'? Considerations of expense (communication equipment) and efficiency. How do 'leader robots' direct the swarm?
  • Command modalities (longer term research) - console, speech, gesture, sound, light shining, clapping, remote controller (IR, sound - dog whistle).


Tie in with my work to date
  • Functional self-assembly is a form of swarm decision making. This ties into HSRI (human swarm-robot interaction) because an essential part of swarm control system design will be to determine to what level the swarm can / should behave autonomously.
  • All my work to date can be used as behavioural components that can be launched by the control mechanisms I will design over the Phd. e.g. Command is : 'Head to that light. If you meet an obstacle functionally self-assemble into groups of size 4'. OR 'Head to that light. If you meet an obstacle get into pairs and adaptively rotate over the hill'.


Shorter Term

  • Finish two sbot hill hill rotation experiment and write it up for a paper.
  • Pattern generation experiment with Anders - parameterised controller that can generate different patterns (chains, circle etc). Write up as a paper for a low key conference. I can use this simple controller as a starting point for investigating swarm control. Anders can use this controller as a starting point to investigate swarm fault detection.


Meeting 2nd December 2005

Long Term - Phd Thesis

Theory
Explore aspects of functional self assembly.
  1. Decision Process
    • Is self-assembly necessary?
  2. Structure
    • What is appropriate size for connected groups?
    • What is appropriate shape for connected groups?
  3. Behaviour
    • What to do when connected (cooperative movement, transport, navigation, etc)?
  4. Timing
    • When should swarm assemble?
    • When should swarm disassemble?
Practice
  1. Parametrised group size selection mechanism (2,3,4,..)
    • Initially in a non-adaptive way
    • Maybe in an adaptive way
  2. Parametrised Shape distinction mechanism
    • Initially binary parameter - linear / blob (non-linear)
    • Initially non-adaptive
    • Maybe adaptive
    • Maybe further parameterisable - control over what 'blob' looks like
  3. Application of above two mechanisms to hill passing problem
    • Comparison of adaptive linear approach and blob approach


Short Term

Timescale

Conference paper writtern by end March 2006

Goal

Conduct new experiment which builds on previous functional self-assembly experiment. Publish new results in conference (probably Ants2006). Combine new conference paper with ecal2005 paper and publish results in journal.

Experiment

Environment
  • Similar to previous ecal2005 functional self assembly experiment.
  • Except use three different hills to prove adaptive rotation mechanism. | / \
Task
  • Two robots must navigate over hill towards target.
  • When they encounter hill, they use adaptive rotation (adaptive to orientation of hill) to avoid toppling.
Methods
  • Reuse existing functional self assembly controller. (Published work)
  • Develop adaptive (to hill orientation) control. (New work)
Trials
  • 2 robots: 20 trials on each hill type. = 60 total.
  • 3 robots (maybe): some trials needed to show blob behaviour doesn't change with different hill orientations.


Misc Ideas

  • Spend 1 or 2 weeks helping Lausanne disassemble / repair robots
    • I get to learn more about robotic hardware
    • Lausanne gets work done faster with extra pair of hands.
    • Iridia gets skill transfer on hardware disassembly / repair.