search

ptsneves / rt-rrt

RT-RRT
GNU General Public License v3.0
11 stars 6 forks source link

Node extracted twice #1

Open jpoikela opened 6 years ago

jpoikela commented 6 years ago

I think you have a bug here: https://github.com/ptsneves/rt-rrt/blob/8f40302f242712f856092ce7d196bc7c347e867b/rt.py#L299 x_closest is already the node of the result

ptsneves commented 6 years ago

I cannot understand this part. I looked at the paper and this seems to be correct. Can you elaborate, or tell me how can I verify this duplication?

jpoikela commented 6 years ago

what I mean is that you already have: x_closest = result[NODE] then you take x_closest[NODE] in the suspicious line which does not seem correct (but may still be)..

ptsneves commented 6 years ago

Will take a loot at it. Also I think this algorithm is quite crappy for real life world discovery. I will try to contact the authors of the paper to see what they think.

jpoikela commented 6 years ago

I wrote a function that draws the routing tree onto the map (with some small changes).. the routing tree on map looks different from the ones shown in the demo videos.. I could push it to github in the evening and make a pull request, I think it gives insight into how the algorithm works..

also there could be space for some speedups here and there..

On Fri, Feb 23, 2018 at 2:08 PM, Paulo Neves notifications@github.com wrote:

Will take a loot at it. Also I think this algorithm is quite crappy for real life world discovery. I will try to contact the authors of the paper to see what they think.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/ptsneves/rt-rrt/issues/1#issuecomment-367992334, or mute the thread https://github.com/notifications/unsubscribe-auth/AAF5cuEMzRSgK2ZS35ZHGkP_skD8rhZwks5tXqongaJpZM4Ru6lR .

ptsneves commented 6 years ago

Sure. I think that a review by the authors, plus better presentation can be beneficial for both of us.

Regarding the ellipse sampling, basically once you find a route to the goal you take the path distance as a parameter for the ellipse for convergent sampling. I was honestly quite disappointed with this step as I was thinking of the ellipse sampling as some mathematical heuristic that could help us get to the goal faster, but it seems it is some form of ad-hoc optimization heuristic.

In the mean time I modified my code to provide the shortest path given the known nodes, basically short-circuiting the graph-tree when there is no obstacle. I am curious if I am not doing some La-palice here :D

jpoikela commented 6 years ago

I agree that the authors would be best guides. shortest routes, nice!