Table of contents
  1. Yuhao
  2. Edison
  3. Abhinav
  4. Zisen
  5. Dmitry, Abhinav & Edison
  6. Adam

Yuhao

  • Fix performance problem in PYLogReader
  • Write the doc for PYLogReader
  • Has collected and parsed data
  • Currently experiencing issues with imitation learning training environment
  • Algorithm crashes periodically during execution

Edison

  • Edison’s brief update:
    • Working with Abhinav on testing AbstractSim
    • Reading and understanding the codebase in his own time
    • Focused on code comprehension
  • Limited progress this week

Abhinav

  • Project status:
    • Coordinating with Zhihan about V2 transfer
    • Planning to train policy in V2’s new branch
    • Will test deployment on SimRobot
  • Performance metrics:
    • CHTC: 1500-2000 FPS
    • Adam Desktop: also 1500 FPS
    • CHTC slower due to single core training limitations ?
  • Current status: Policy training ongoing

Zisen

  • Multiple workstreams:
    1. PyLogReader:
      • Yuhao has completed debugging
      • Not tested yet by Zisen
    2. Research:
      • Reading imitation learning literature
    3. Referee Gesture Recognition:
      • 99% accuracy on training and validation
      • Model inference time: 0.2 seconds
        • Adam ask if Zisen use the network of BHuman and make a reminder that the network need to be small enough to be run on Nao
      • Considering architecture options:
        • Build CNN from scratch
        • Use transfer learning with ImageNet
  • Competition Rules Discussion:
    • Referee uses whistle signal
    • Robot moves to field center
    • 12 different gestures to recognize
    • Robot outputs sound for recognized gesture

Dmitry, Abhinav & Edison

  • Setup Issues & Solutions:
    1. AbstractSim V2 Problems:
      • CV2 import errors
      • Missing libGL package
    2. Solutions provided:
      • Comment out CV2 import line
      • Install libGL package on WSL
      • Use conda.yaml file
  • Environment Setup:
    • Edison confirms conda environment works on Mac with modifications
    • Abhinav recommends using provided conda environment to avoid issues

Adam

  • Mujoco Project Status:
    • No significant progress
    • Pending RL environment integration
  • Bottlenecks:
    • Time constraints ?
  • Open invitation for collaboration