Progress for week 3 (2020)

Fra Robin

(Forskjeller mellom versjoner)
Gå til: navigasjon, søk
(Ny side: == Nikolai == === Accounting === === Budget === https://youtu.be/-_4LFfClEzE)
(Wonho)
 
(Én mellomrevisjon ikke vist.)
Linje 4: Linje 4:
=== Budget ===
=== Budget ===
https://youtu.be/-_4LFfClEzE
https://youtu.be/-_4LFfClEzE
 +
 +
== Wonho ==
 +
=== Accounting ===
 +
* Somewhat stable gait evolved in static, and some plausible one in flexible - [[https://robin.wiki.ifi.uio.no/Bruker:Wonhol#Week_3]]
 +
* novelty search is implemented by comparing history of each joint against every other genome's behaviour
 +
* parallel processing for novelty search - recording behaviour + comparing. note that this is two step operation
 +
* have been implementing script for Abel - currently both multiprocessing and MPI doesn't work on SLURM
 +
=== Budget ===
 +
* Need more computing power - currently flexible controller takes 50+ hours on 4 core machine
 +
* Implement Novelty search based on position history of the agent as Risi suggested - contrary to comparing each joints
 +
** This will take more time on recording behaviour - currently recording only the first 200 iteration, which is couple of period on ctrnn output
 +
* Find out which of simple CPG based controller can be implemented to compare with the current implementation.

Nåværende revisjon fra 15. jan 2020 kl. 21:05

Innhold

Nikolai

Accounting

Budget

https://youtu.be/-_4LFfClEzE

Wonho

Accounting

  • Somewhat stable gait evolved in static, and some plausible one in flexible - [[1]]
  • novelty search is implemented by comparing history of each joint against every other genome's behaviour
  • parallel processing for novelty search - recording behaviour + comparing. note that this is two step operation
  • have been implementing script for Abel - currently both multiprocessing and MPI doesn't work on SLURM

Budget

  • Need more computing power - currently flexible controller takes 50+ hours on 4 core machine
  • Implement Novelty search based on position history of the agent as Risi suggested - contrary to comparing each joints
    • This will take more time on recording behaviour - currently recording only the first 200 iteration, which is couple of period on ctrnn output
  • Find out which of simple CPG based controller can be implemented to compare with the current implementation.
Personlige verktøy