Yuchen Wu

The path tracker should not look past the cusp until the robot is close enough to it.

Log, screen recording, and issue explanation here: https://drive.google.com/drive/u/2/folders/12fKPxI9GqUSkbG8k8EYjXrJUln-Pewg_

I've always assumed that the MEL feature matching visualization happens in the mel_matcher_module with an enabling flag: https://github.com/utiasASRL/vtr3/blob/3eb447d12d9349964358c4a7a62033af82311570/main/src/vtr_vision/src/modules/matching/mel_matcher_module.cpp#L54 However, the visualization function is actually called in ransac_module here: https://github.com/utiasASRL/vtr3/blob/aae397c80642b8bc3e6089342ede4e085db9dba3/main/src/vtr_vision/src/modules/ransac/ransac_module.cpp#L202 Does...

When user confirms a merge point by clicking on the yellow triangle, internally we check if the distance to the merging vertex is too large to merge. However, if a...

Force to metric-localization state during repeat does not work as expected (after recent cusp issue fix on path tracker side) - disable this function for now in the UI (see...

See: https://drive.google.com/drive/folders/1YvgF9iwTWB3pvqntZ0Qge8zB9TxKTzhk?usp=sharing videos and log file

So far this may happen after a long (likely >2km) repeat for the camera TR case.

Add instructions or change the selector. [T1-3:52] https://drive.google.com/file/d/1ictclQfTq7oL9txlkxj6qh3lUpIeQlU-/view?usp=sharing

The [RTR paper](https://arxiv.org/pdf/2203.10174.pdf) assumes access to the ground-truth relative transformation of the first frame for initialization. For a fair comparison, consider adding a note to the localization benchmark saying using...

https://github.com/utiasASRL/pyboreas/blob/7a133355302a35a3e2c328a9d3d7ae27df188d7f/pyboreas/utils/odometry.py#L737 Either pitch or roll should be pi instead of 0 - maybe just round line[8] and line[7] to the nearest k * pi?