Question: I recently opened github.com/ros-navigation/navigation2.ai. Would parts of this fall into that? The costmap layer I know we've talked about (maybe submodule). But is the node sufficiently generic? It seems so.

I might have a few more scrutiny comments on it for efficiency if adding to github.com/ros-navigation/navigation2.ai, but if its general (i.e. any ONNX file works) then I think that would be a good not-just-tutorial contribution :-)

Question: I recently opened github.com/ros-navigation/navigation2.ai. Would parts of this fall into that? The costmap layer I know we've talked about (maybe submodule). But is the node sufficiently generic? It seems so.

I might have a few more scrutiny comments on it for efficiency if adding to github.com/ros-navigation/navigation2.ai, but if its general (i.e. any ONNX file works) then I think that would be a good not-just-tutorial contribution :-)

inference should work with any onnx file, though I think the input, output and confidence handling may not be the same for all models (ex: ultralytics YOLO), or users may not wanna use the max rule for selecting the confidence. For isntance, this node applies imagenet normalization to the input, but I'm not sure that all segmentation models would produce correct results with this. Do you have any other model you think users may wanna run?

TBH I'd leave the node in the tutorial for now

Ok :-)

thanks for your review! i think i adressed most comments in both PRs