Fix A* star selection logic #22
Conversation
Schmluk
left a comment
Schmluk
left a comment
There was a problem hiding this comment.
LGTM, nice catch! Only one minor comment.
| (t_odom_current_vertex - start_point).norm(); | ||
| f_score_map[current_vertex_id] = | ||
| (goal_point - t_odom_current_vertex).norm(); | ||
| f_score_map[current_vertex_id] = h_score; |
There was a problem hiding this comment.
Is this intentionally called h_score and then assigned to the f_score_map? I'm not too familiar with the terminology but might lead to confusions.
There was a problem hiding this comment.
Strictly speaking, the update equation here should be f[neighbor] = g[neighbor] + h[neighbor]. The code above only initializes the starting vertices and will result in all of them being expanded before any regular vertex, so it won't affect the result. But I agree it's confusing and there's no advantage of not using the standard update equation.
In general, this whole getPathBetweenVertices(...) method grew organically and got quite messy. I'll do some refactoring.
…ludes Automatically order includes with clang-format
Update the GLocal ArXiv link to the right paper
Update README.md
Update README.md
This PR fixes A*'s best-path selection comparison, which was wrongly inverted.
Unsurprisingly this significantly improves performance, especially when the linked skeleton graph becomes large.
This bug also affects
mav_voxblox_planningand the other packages in our stack that are derived from it. Corresponding PRs will be opened there.