4.14 Bug Report Spline Mesh Collision Corruption

When drawing out a long spline mesh ( railway track) after a certain distance the collision mesh becomes corrupted.
In 4.14

IN 4.12

This has been repeated in multiple vanilla projects, using entirely different BP’s

The collision mesh seems to lose a vertex after a set distance from the origin.

(in my cases it seems to be around 360 m ).

The distance seems to be affected by the cross sectional area to length of segmant.

(it does not appear to be affected by just the length of each segmant).

It affects both Custom UCX collision and collision made by adding boxes in the UE4 SM editor.

I have checked everything I can find regarding LOD

(all SM’s have base LOD only)

Simplest way to repeat, I use the simle spline mesh tutorial by Dan Elliot (Dokipen).

Make a spline with cube as mesh, drag out a long spline
check collision … as mentioned mine happens around 360 m

Any Ideas ? or is this a bug in 4.14
( have been using this method without problems 4.7 thro 4.13

I can supply a small project with the errors if req

Paul G

Hi Sly401,

I set up something simple in a blank project but I was unable to reproduce this behavior. If you could upload the test project you mentioned that would be helpful.


Thanks for takin a look

I have a tiny project readt for upload,
I cannot get the size lower than 11.8 mb which exceeds the upload limit.

Is there an email or upload link I can send it to ?

Notes on the Project

3 seperate splines laid out
Switch to player collision view mode to see the corruption

2 are using Dan Elliots tutorial method and have a stock 1m cube as the mesh.
the 1m cube has had its collision replaced with a box that has a different LWH ratio
two examles are to demonstrate that the repeat pitch seems to make little or no difference

The final example is using a more complex script that involves redrawing the track after laying out the spline.
The custom track mesh uses a series of boxes added for collision
(this track used to have a UCX custom collision… the results are were identical)

The ratio of width / height / length makes a difference.
I seem to recall that a 1m cube with a 1mcube collision did not exhibit this behaviour.


Paul G

Try my dropbox link for the projrct


Hi Sly401,

Sorry for the delay and thank you for the test project. I was able to reproduce this and it is indeed a regression from 4.13. I’ve created JIRA UE-39212 and our developers will be investigating further. Follow that link to monitor the status of the bug report and please feel free to vote.




Thanks for the confirmation, glad to know its not just me, have voted and look forward to being able to use some of the excellent 4.14 goodies

Paul G