Tasks 26-27-28-29

[mmattamala]

Immediate post-trial tasks

• [ ] Fetch all the data from:
  • [x] Robot
  • [x] GoPro
  • [ ] Canon camera
  • [x] Make a tag for the changes we used on the robot

Analysis

• [ ] Inspect the logs:
  • [ ] Check how learning looks like
  • [x] 1. Confidence & Traversability -> FPR threshold collapsed
  • [ ] 2. Debug the learning curves (why do we get 0 learning at some point)
  • [ ] How elevation mapping looks like
  • [x] Rework the visualization of the NANs
  • [x] Compare geometry costmap vs traversability cost map (Create a plot by changing the alpha value for visual, geometrical and SDF visual and SDF geometry)
  • [x] Analyze impact of false positive predictions on elevation map / cost maps
  • [ ] Check local planner behavior
  • [x] How do the costs look like
  • [ ] Integrate the GDF to get a path and publish it
  • [x] Analyze was there a problem that the locomotion policy did not follow the commanded velocity by the local planner (I think yes) -> Traversability information was noisy
  • [ ] Mission report : link

Paper writing

• [ ] Determine what's good to put on the paper

Showing system capable of adaptation

• [ ] Loop experiments from the Oxford Park
• [ ] Show how the predictions change / on image space and maybe map level
• [ ] Check how the loss is changing over time

Closed-loop integration done today

• [ ] Look at all the results
• [ ] Come up with a proper way of framing

Traversability Method Comparision

• [ ] High grass (oxford park and forest) ( Geometrical vs Visual Traversability )
• [ ] WayFast ?

[mmattamala]

Local planner

• test if a goal in front biased towards left/ right works
• Test RMP local planner in differential mode
• Inflate obstacles at the traversability level
• Test FALCO
• Implement simple setup in simulation