Skip to content

2019_11_08

Gabriel Perdue edited this page Nov 8, 2019 · 1 revision

LDRD Discussion 2019-11-08

Meeting Notes

  • Discuss possible dates and-or time windows for an onsite meeting.
    • Brian is broadly available
    • Jason is broadly available
    • Christian is traveling in early December (8th - 10th, suspect Nhan will be there also)
    • Malachi traveling 1st week of December?
      • Maybe we'll try schedule second week of December
        • Jason recommends when2meet
  • Algorithm idea - Lucy's idea to tune constants in current algorithm
    • first supervised (fit the constants), then RL (tune the "constants" on the fly)?
    • two board model
      • old board
      • new board with exact same code -> then moves to ML
        • might also put both algorithms on new board if there is space
        • should be able to fit and would offer some nice comparison options
    • Lucy showed some interesting plots of behavior - updating proportional and integral gains
  • follow up
    • zero gain runs
      • couple of hours of data
      • running power supply open loop
      • constants in control alg. should have been zero - if true, what is our error? turns out to be quite large
      • data access issues?
        • on tape? - need to stage?
          • multiple people having trouble writing to home area on gpvm
      • Kiyomi is studying the data
    • ferry on-boarding
  • GitHib project roundup
    • data products
      • automated archival with group account is not working, but is being worked on
    • low latency data infrastructure
      • recap conversation with Erik Gottschalk
      • how do we get more plugged in here?
      • 30,000 foot level
        • Bill and Brian had a conversation - part of the modernization effort
        • Big task that will take many years, trying to understand all the impacts for ML
        • Hope to build a universal system
        • Data logging and collection is tough - this needs to be fixed for us to make progress
        • We should stop thinking about piggy-backing on efforts in AD modernization, time-scales are out of sync here
          • However, we are in a position to offer interesting feedback and express our needs - might help in design of future systems, etc.
    • board deployment
      • all of the software is installed on local machine (not piggy-backing off of Nhan)
      • 80% of the way there
      • board is running, can connect to it as a Linux machine
      • still a borrowed board, ours should arrive next week
      • trying to get PCB layout for motherboard, then we will make edits and changes specific to this project and send a new revision
      • global system layer? - we aren't actually worried about this on this scale
      • working on digitizing the transductor system - not working yet... board next week does not have ADCs, but Philip's board does
        • our intermediate board cannot do this
        • dev kit has no digitizers - we need final board
      • timeline for final board layout
        • universal board for AD for ML and some other things
        • layouts take 3 weeks to complete - start soon
        • then once the order goes to the vendor, it should take less than 6 weeks
          • maybe get final board by March 1st
Clone this wiki locally