It should do but it's a computer
it should do
because it's a computer
They're not random, they haven't got a mind of their own. They perform calculations based on a set of rules, which are predetermined - in routing, those rules are that there is a matrix of "nodes", each of which is connected to a number of adjacent nodes, where each connection has a 'cost' associated with it, and it routes by walking from a start node to every adjacent node, and from each one of them to each of its adjacent node, and so on... until it gets to the target node.
Obviously it also prunes certain branches, as an optimization, as to calculate the cost of all the permutations would take forever, but essentially I simply do not see how the number of choices for connecting two points which only have a couple or three nodes between them can yield enough similar routes for there not to be one obvious one that stands out. Which is the 'correct' one, i.e. the one that it's obvious to a human has the least cost. In other words, it should be possible to feed points into it that is friendly to the way
it works, such that it is obvious to it how to pick the route that
we humans consider the correct one.
Certainly the router we have got at work would do that, and I can't imagine garmin would have a worse one that some small backstreet company like us, so it must be something else - I suspect to do with the translation from the lat long of the point you give it to the node/connection that it considers that to be part of.
A translation which, remember, is a non-trivial step since the connections (roads) are as far as the algorithm is concerned, the width of a mathematical line - in other words, infinetessimally thin - so it must assume it to be on the one that is the smallest eucldean distance away. No need for fuzzy logic there, but the only place I can see where it could go wrong if it is calculating the costs between nodes correctly.
It's also possible that with the unit probably running low-power cmos, the pruning is doing something crazy. This is obviously controlled by the 'best/better/quicker/quickest' setting, but i wonder if even on best, it is still a bit arbitrary?
, it won't work properly, they never do. (Moi, cynical?) And anyway if you have to put in so many points, you've basically got a Track!
I agree that the confidence that you gain from not allowing its algorithm to mess you up is an overpowering reason to always use tracks. The only thing you don't get is the "distance to turn" countdown, which is the main reason i like the turn by turn mode.