Meeting Updates

[mmattamala]

Agenda 1

1. Neglect the GPS most likely.
2. Baselines
   • WayFast try to retrain it and inference exactly on our data. See the visual difference.
   • Using Geometrical Traversability
3. Regenerating the ablations and making the plots.
4. Calibrations Camera Lidar Perugia
   • /media/Data/Datasets/2022_Perugia/day3/mission_data/2022-05-12T11-44-56_mission_0_day_3�
   • Make sure to use correct alphasense_rsl branch
5. Current Timings:

   Image callback: front... process
   Time image_callback - preprocess:  21.038463592529297 ms
   Time feature_extractor - sparsify_features:  51.83488082885742 ms
   Time add_mission_node - update_features:  266.98651123046875 ms
   Time add_mission_node - optical flow:  39.82950210571289 ms
   Time add_mission_node - project_supervision:  9.16096019744873 ms
   Time add_mission_node - update_supervision_signal:  1.742624044418335 ms
   Time image_callback - add_mission_node:  345.8580627441406 ms
   step: 880 | loss: 1.521120 | loss_trav: 0.169302 | loss_reco: 1.372939
   Time publish_mission_graph:  498.3157958984375 ms
   Time visualizer - plot_traversability_graph_on_seg trav:  147.77203369140625 ms
   Time visualizer - plot_traversability_graph_on_seg conf:  148.0391387939453 ms
   Time plot_mission_node_prediction:  298.7796630859375 ms
   Time publish_mission_node_prediction:  0.886784017086029 ms
   Time plot_mission_node_training:  359.99627685546875 ms
   Time publish_mission_node_training:  0.9010879993438721 ms
   Time image_callback - update visualizations:  1202.3193359375 ms

6. Integrate the output into RMP

Agenda 0

Current status of the system

@JonasFrey96

• Elevation mapping: https://github.com/leggedrobotics/elevation_mapping_semantic_cupy/pull/26
• Ablations:

@mmattamala

• Robot integration (pre-Christmas): https://drive.google.com/drive/folders/1ua8b4URPF76ZnvQAkgRkEEzhJNGEs9Xb?usp=share_link
• Local planner refactoring: New branch: https://github.com/ori-drs/locally_reactive_controller/tree/dev/refactor-field-local-planners

Pending tasks

• Open issues - mainly to improve performance on robot

  • 150

  • 151 that solves some of those things)

  • 132 These summarizes other things that need to be done

• Analysis of non-conclusive ablations Just by solving these we should be able to fill the paper and have something reasonable

Oxford visit

• Flight tickets -> ready
• Accommodation -> ready
• Airport to Oxford -> The Airline simplest choice: https://www.theairlineoxford.co.uk/
• Time of arrival UK? Arrival to Oxford?
• Desk: should allocate one next to mine (Lintong is in China)
• Falafel :falafel:
• Experiments:
  • Need to arrange days to go to the park beforehand because of safety
  • Potential places for experiments:
  • University parks is a decent spot: https://goo.gl/maps/4RRQFycVu6wrxMBz6
  • Port Meadow / Burgess Field may work too: https://goo.gl/maps/zTyao48BowU4A6577

Other things to do

• Orin
  • I can take a look when I return to Oxford next week.
  • Also prepare the Docker container
  • Not sure about installing it decently on the robot

[JonasFrey96]

My flight:

[JonasFrey96]

• Check the logs of the Oxford experiments and look through them.
• Run Multiple Camera PR
• Bring Battery
• Review Gian PR Elevation Mapping
• Talk to Gabriel Turcan and Lorenz about GPS
• Start-Stoping-Training
• Rerun the ablation studies we discussed
• Send notebook with runnable ablation studies
• Send PR for Abblations

[mmattamala]

• Share logs with Jonas via RSL folders
• Ask about GPS

[JonasFrey96]

@mmattamala I don't have to opportunity to bring our full GPS unit but I could set up a new unit. I think we have everything for this available but it would require some work (4-8hours). Should I do this ?

[mmattamala]

Let me ask if we received the components we were missing first, I'll keep you posted