Shortest path on recast graph

I ran into a problem where after runtime graph updates an enemy agent starting taking the long way around instead going directly ahead at their target.
I searched the forums here and a couple things came up which helped, turning on funnel simplification on RichAI, and lowering the max edge length (default was 20, lowered it to 10).

This only showed up due to how I was laying out my walls (which caused cuts to the navmesh and graph updates).

Are those 2 options the best plan? I counted the triangles the agent decided to use and it seems the long way they chose is 20, and the direct path is 14 so I don’t see why they chose the long way?

Also, why would you ever then NOT want to use funnel simplification?

2 images, one without any notes so you can see clearly:


In the beta version, I’ve made improvements to how pathfinding on navmesh-like graphs is done. Pathfinding is much more accurate now. So if you can, check out the beta.
See A* Pathfinding Project

Improved accuracy of path searches on navmesh graphs significantly.
    Previously, paths on navmesh/recast graphs have been searched by going from triangle center to triangle center. This is not very accurate and can lead to agents choosing suboptimal paths. Now, paths use a more accurate method which moves between different points on the sides of the triangles instead of the centers. This is much more accurate and should lead to agents choosing better paths in almost all cases.
    It also accounts for the exact start and end point of the path more accurately now.
    However, this improved accuracy comes with a cost, and pathfinding performance on navmesh/recast graphs is a bit lower than before. Luckily, it is still very fast, and most games are not bottlenecked by this.

What is your opinion on using the beta in a game that is about to released?

It’s just a fairly simple tower defense so I’m not doing much beyond using multiple recast graphs for different agent sizes, and runtime graph updates for tower construction etc.

The beta is pretty stable overall. Things sometimes break in the beta, but I try to release fixes for those issues pretty quickly. I’d say try it out and see if it works for you.