Progress for week 10 (2020)

Fra Robin

(Forskjeller mellom versjoner)
Gå til: navigasjon, søk
(Budget)
 
(3 mellomrevisjoner ikke vist.)
Linje 5: Linje 5:
* Hvordan logge?
* Hvordan logge?
* Trenger flere folk til 2.april
* Trenger flere folk til 2.april
 +
== Wonho ==
== Wonho ==
=== Accounting ===
=== Accounting ===
* Running experiments, getting results
* Running experiments, getting results
 +
** Most of planned experiments either completed/running or on the queue
* behavior descriptor results
* behavior descriptor results
-
[[Fil:Beh_descriptor_MBF.png|500px]]
+
[[Fil:Beh_descriptor_MBF.png|300px]]
-
[[Fil:Beh_descriptor_MBF_GEN_ALL.png|500px]]
+
[[Fil:Beh_descriptor_MBF_GEN_ALL.png|300px]]
-
[[Fil:Beh_descriptor_ns.png|500px]]
+
[[Fil:Beh_descriptor_ns.png|300px]]
 +
 
 +
 
[[Fil:Anova_mbf.png]]
[[Fil:Anova_mbf.png]]
[[Fil:Pairwise_ttest_mbf.png]]
[[Fil:Pairwise_ttest_mbf.png]]
 +
 +
(Are these okay to report as such?)
 +
* NEAT MOEA experiment turned out to be somewhat flawed
* NEAT MOEA experiment turned out to be somewhat flawed
** MBF higher than hyperNEAT but most genomes are exploiting simulator - they are all gliding around.
** MBF higher than hyperNEAT but most genomes are exploiting simulator - they are all gliding around.
-
** How to show this in paper & how to provide analysis? - trajectory of foot patterns?
+
[[Fil:Neat_gliding.gif|200px]]
 +
** How to show this in paper & how to provide analysis? - 2d trajectory of foot patterns?
=== Budget ===
=== Budget ===
 +
* Need to plan how static vs flexible comparison should be done - the controllers are evolved
 +
** follow risi by averaging and comparing distance traveled on trained leg configs
 +
** compare interpolation performace - leg configuration not seen during training
 +
* Continue writing implementation
 +
** how to credit Joergen's DyRET openai environment?
== Tony ==
== Tony ==
Linje 27: Linje 40:
=== Budget ===
=== Budget ===
* Skal få opp resultater på en god format.
* Skal få opp resultater på en god format.
 +
 +
 +
 +
== Daniel ==
 +
=== Accounting ===
 +
* Implementerte prototype for første eksperiment (diskrete noder) i NEAT. Restultater under:
 +
[[Fil:gamesize-64-numgames-32-generations-300-2802-digraph.png|300px]]
 +
[[Fil:gamesize-64-numgames-32-generations-300-2802-fitness-generations-plots.png|300px]]
 +
[[Fil:gamesize-64-numgames-32-generations-300-2802-speciations.png|300px]]
 +
 +
=== Budget ===
 +
* Hvordan fungerer "recurrent nodes" i NEAT?
 +
* Hvordan gjøre spillet mindre "random"?
 +
* Implementere egne aktiveringsfunksjoner i NEAT.
 +
* Implementere prototype for andre eksperiment (kontinuerlige noder) i NEAT.

Nåværende revisjon fra 5. mar 2020 kl. 10:42

Innhold

Nikolai

  • Trenger hjelp til å vite hva som skal med i Background
    • Vanskelig å skrive komprimert om SLAM
  • Hvordan logge?
  • Trenger flere folk til 2.april


Wonho

Accounting

  • Running experiments, getting results
    • Most of planned experiments either completed/running or on the queue
  • behavior descriptor results


Fil:Anova_mbf.png Fil:Pairwise_ttest_mbf.png

(Are these okay to report as such?)

  • NEAT MOEA experiment turned out to be somewhat flawed
    • MBF higher than hyperNEAT but most genomes are exploiting simulator - they are all gliding around.

    • How to show this in paper & how to provide analysis? - 2d trajectory of foot patterns?

Budget

  • Need to plan how static vs flexible comparison should be done - the controllers are evolved
    • follow risi by averaging and comparing distance traveled on trained leg configs
    • compare interpolation performace - leg configuration not seen during training
  • Continue writing implementation
    • how to credit Joergen's DyRET openai environment?

Tony

Accounting

  • Har fokusert mest på skriving for denne uken --> Motion-Capture og Genetisk Algoritme.

Budget

  • Skal få opp resultater på en god format.


Daniel

Accounting

  • Implementerte prototype for første eksperiment (diskrete noder) i NEAT. Restultater under:

Budget

  • Hvordan fungerer "recurrent nodes" i NEAT?
  • Hvordan gjøre spillet mindre "random"?
  • Implementere egne aktiveringsfunksjoner i NEAT.
  • Implementere prototype for andre eksperiment (kontinuerlige noder) i NEAT.
Personlige verktøy