Progress Review 4

Localization

  • Got agents to connect to CMU-DEVICE WiFi, enabling testing outside VICON area
    • Borrowed WiFi router from lab to avoid network issues on CMU-DEVICE
  • Tested robots in NSH basement and FRC
    • Relocated testing to Scott Hall, which has polished floors
  • Generated map of Scott Hall and tested localization
    • Recorded data and tuned localization parameters
    • Localization now works reliably

Task Allocation

  • Improved optimality: avg 28.8% lower objective value than greedy
    • Downsampled map occupancy grid to for faster A*
    • Fully connected graph for more accurate costs
    • TSP -> Hamiltonian path formulation
    • Increased time for local search with metaheuristics
  • Created fleet update feature
    • Recomputes optimal allocation if fleet size changes
  • Integrated with system
    • Update active agents and tasks via service call
    • Get all map information from ROS topic

Path Planning

  • Spatio-Temporal A*
    • Prioritized planning for each robot
    • Planning in time to avoid robot-robot collision
  • LG Control
    • Custom controller to follow global trajectory

https://drive.google.com/file/d/16bkaGS6nz5thRhFPEA0g9kCvj7MPV8tR/view?usp=sharing

Task Generation

  • Tested out art gallery formulation
    • Doesn’t seem to be most suitable for our applications, will be pivoting to Voronoi diagrams