wiki:Carologistics/Protocol/2018-07-11

Version 7 (modified by tviehmann, 2 months ago) (diff)

--

Caros Team Meeting 2018-07-11

Agenda

RoboCup Retrospective

  • what did go well:
  • progress in the agent looks like stable base for further development
  • real life testing of sync + lock
  • experience with wifi issues
  • work motivation was great
  • clearly distributed tasks at RoboCup through smaller subteams
  • what did go not well:
  • missing motivation/time before the RoboCup
  • wifi issues, which we were not prepared for
  • getting team meetings more frequently (before RoboCup)
  • not enough time spent on preparation + testing
    • no proper testing possible beforehand
    • no proper field
    • key components still under construction
  • Suggestion: team day once a week with team meeting
  • getting everyone together
  • give reason if u cant attend and also why
  • Get information about time schedules from everyone in the new future
    • give information when you cannot work in certain timespans
    • participate in meetings via skype/phone
    • spread responsibilities
      • Have one person responsible for each mayor componement (be aware of tasks and progress, schedule work)
    • more definite commitments, frequent updates
  • Technical challenge
  • Not so important compared to the real competition
  • splitting the team when we worked at night
  • while wifi issues/storage station in the beginning agents had problems working
    • simulation was impossible
  • task: get new mongodb driver integrated
  • General Remarks:
    • Expect hardware failures anywhere (even with the access points)
    • Do not do wifi bridging anymore
    • still have spare wifi hardware
    • suggestion: PCI cards for wifi, laptop as fallbacks
    • Get a PCI card that supports enough frequencies
    • Suggestion: proper task distribution (especially after german open) to keep everyone

motivated

  • find task for yourself but communicate everything
  • New members got involved quickly
    • not new members for technical challenge only
    • To integrate new people build pairs
  • Wunderlist for merging stuff
    • create git plugin to mark branches to merge?
    • Problems with setting up in time
    • SSD on the robots seem to be broken
    • Review all the hardware componements
    • replace old hardware
    • More tool support (Migrate fawkes git to GitLab? or similar)?
      • pull requests
      • code reviews
      • issue tracking
    • Ease up operator tasks
  • State of the other teams:
    • Pyro has stable + fast gripping, better movebase, state machine issues
    • GRIPS improved gripping
    • ErForce? starts producing
    • Solidus moved in the production phase, not producing yet
  • Gripping:
    • hardcoding conveyer offsets relative to machine
      • not possible by using lasers because endpoints jitter
      • tag camera to align relative to tag
      • could work, we do not like it
    • Overengineered approach currently?
      • measure position of conveyer belt relative to belt and make hybrid approach
      • currently ICP models are a bottleneck
      • model for every side of every machine is too much work
      • discard vision approach in favor of simple lasers?
    • Old Gripper is broken
    • Rework the Gripper
      • Suck the workpiece to the gripper instead of parallel gripping?
      • Discuss with David
  • Vision:
    • add more debug functionalities
      • save ICP fits to review later (including timestamp + name of machine)
    • Review all Gripping Skills
      • MPS alignment as plugin instead of skills?
      • Fix parallel skill execution?
    • No ICP but keep RGB without depth information?
      • depth information is unreliable in the area of the belt
      • was already done before, was not precise enough
      • use modern Computer Vision
        • z information?
        • retraining networks costly
        • rotation?
    • generic image recognition integration
      • workpiece detection
  • Agent
    • New approach allowed easy debugging
    • Tasks
      • Improve the Webview
      • have single agent fallback
      • Debug the locking
      • Reinsertion of a bot
      • not enough information to properly reason
        • need reliable gripper + workpiece information
      • gain more knowledge about locking/mongodb and all agent components
    • Exploration
      • keep locking+syncing
      • adapt decision on the zone to explore for syncing
      • block zones where mps are not allowed due to rules
  • Navigation
    • Issue of faulty TF publishing that leads to the planner getting stuck
      • Conveyer Pose should not publish faulty TFs
      • TF code is difficult to debug (not important)
    • Collisions that occured are fixed (caused by velocity smoother)
    • Next year: local planner based on potential fields
      • needs a lot of tuning
      • should be more efficient
    • distinct laser beams coming from static and dynamic obstacles
    • Multi-Agent Navigation
      • centralized approach
      • potential field based approach
        • include other enviromental information
      • hard coded priorities for conflict resolution
      • NavGraph based soft blocking of edges
        • low NavGraph point resulution leads to inefficient pathes

  • AirBnB? Travel?

Planning the next year

  • Commitment
    • David:
      • next semester is last one as student
      • master thesis on kbsg
      • RoboCup next year not guaranteed, but expected
    • Tarik
      • next semester is last one as student
      • master thesis on kbsg
      • RoboCup next year not guaranteed, but expected
      • gone for from 30.7. to 4.8.
    • Christoph
      • exams until october
      • then one day per week
      • one/two month before RoboCup more time
    • Mostafa
      • planning to finish thesis in october (very optimistic, at letest in december)
      • RoboCup next year is unclear
      • Updates later this year
    • Nick
    • Till
      • until october: occupied with other tasks, still working on clips-executive
      • gone from october to january
    • David
      • on board for RoboCup 2019 (info by Daniel)
      • same time commitment as before
    • Morian
      • probably on board for RoboCup 2019 (then beginning his master studies)
    • Victor
      • same as last year
  • Team Component meeting in the near future 1) gripper 2) vision 3) clips-executive
    • Team Meeting end of September
    • Team Meeting in February
  • Major ToDos
  • new field
    • Robotino 4
  • schedule
  • responsibilities
    • Daniel: vision for additional sensing, more feedback for execution monitoring
    • Till: Replace mongo-cxx driver
    • Tarik: Support single agent fallbacks
    • Victor:
      • Conveyor Pose
      • handing over tasks + knowledge
    • Victor + Christoph: Refactor Conveyor Pose
      • wrap ICP into plugin?
    • Christoph:
      • adapt simulation for storage station and new skills
      • simulate package loss
    • Nick:
      • Gripper
        • faster axis alignment
        • Simplify Gripping
      • Navigation
        • Potential Field approach
    • Mostafa
      • Goal Trees
      • Look into Mongodb
  • Hackathon?