6

I used osm2po to create a table for pgRouting. I notice a problem: for a route with dijkstra or a* the route is correct and respect the one ways. But with shooting*, it doesn't respect the one ways.

The route is from 11 baulacre geneve (46.2159082, 6.1394522) to coq inde cologny (46.2068106, 6.1919172).

Here are the queries: dijkstra/a*:

SELECT topo.id, topo.osm_name, topo.geom_way
FROM Shortest_Path('SELECT * FROM ch_2po_4pgr', 263951, 271357, true, true)
AS path, ch_2po_4pgr AS topo
WHERE path.edge_id = topo.id;

shooting*:

SELECT topo.id, topo.osm_name, topo.geom_way
FROM Shortest_Path_shooting_star(
   'SELECT *, geom_way as rule, reverse_cost as to_cost FROM ch_2po_4pgr', 321089, 309056, true, true)
AS path, ch_2po_4pgr AS topo
WHERE path.edge_id = topo.id;

Did you notice the same problem with shooting* in others situations?

Here is a picture. The orange line is for a*. The red line is for shooting*. Since with shooting* with use edge_id instead vertex_id, maybe the algo starts at the beginning of the way (starting point is in pink), but the problem with the one ways is in the pink circle: the red line uses the "rue des Alpes" which is a one way.

enter image description here

Thanks to all for your help!

underdark
  • 84,148
  • 21
  • 231
  • 413
Michaël Mathieu
  • 61
  • 1
  • from your query, it looks like reverse_cost and to_cost have the same value. – Allan Adair Sep 07 '11 at 12:06
  • if I set cost for to_cost, I have the same problem :s http://imageshack.us/f/263/costfortocost.png/ – Michaël Mathieu Sep 07 '11 at 12:46
  • 2
    For one-way segments, it's a good idea to have a maxed-out reverse_cost assuming that the one-way segment is digitized in the same direction that the edge should be traversed. This will effectively restrict traversal in the wrong direction. There should be both a to_cost and a reverse_cost. – Allan Adair Sep 07 '11 at 13:13
  • I agree :) but it's strange that with the same data set, with a* it works, and with shooting* it doesn't. Anyone have a proposition to modify my query to fix the problem? – Michaël Mathieu Sep 07 '11 at 13:37
  • Can you post some representative rows for bi-directional and one-way segments? I'd like to see what "Select *" is returning and the values for reverse_cost and to_cost. – Allan Adair Sep 07 '11 at 13:58
  • For the one way, the reverse_cost is already huge (first a normal way, and then a one-way): (street name; cost; reverse_cost): "Rue du Vidollet";0.00229477;0.00229477 "Rue des Alpes";0.0011679812;1000000

    Thank you Allan for your interest! :)

     – Michaël Mathieu Sep 07 '11 at 14:06
  • This might just be personal preference, but I would avoid using "SELECT *" within the shortest_path_shooting_star() function. Instead I would suggest explicitly selecting the column names that you need for the function to perform. – Allan Adair Sep 07 '11 at 14:37
  • It was just for a quick test, but I agree. – Michaël Mathieu Sep 07 '11 at 14:39
  • according to research its true....!!!!!! –  Aug 11 '14 at 14:09

1 Answers1

1

Just some ideas:

As far as I remember, the last two parameters mean "has_reverse_cost" and "bidirected". Don't know which one came first ;-( It means, that (if I'm not totally wrong) "true, true" expects a graph where each segment has to occur twice, except for oneways. Another reason might be the nature of the edge-based algo nature. It routes from segment to segment and not from vertex to vertex. As you have already found out, the traversing of the first edge might be a problem here.

whuber
  • 69,783
  • 15
  • 186
  • 281
Carsten
  • 11
  • 1
  • The last two parameters should be "directed" and "has_reverse_cost". Setting it to "true, true" should allow it to establish direction and apply a reverse cost if the edge is traversed in the opposite direction. Here is a link to the docs (not always a great resource): http://www.pgrouting.org/docs/1.x/shooting.html – Allan Adair Sep 07 '11 at 13:55
  • Thank you Carsten, I join you with your second idea: the problem comes from the nature of the algo. I think that we need to use an algo on vertex based. – Michaël Mathieu Sep 07 '11 at 14:00
  • sorry, I meant "directed". I'm not sure which is the correct translation but I assume it means that segments represent directed (one arrow) edges. Setting this to false means it will be traversed in both directions. – Carsten Sep 07 '11 at 14:11