This article discusses the pitfalls of bot navigation, and describes the human approach to spacial awareness and terrain recognition. Base on this, a modular system is described allowing both reactive and deliberative components to cooperate eleguantly.
You can find the essay right here.

I have a slightly buggy 2d game with triangles to represent the avatar and the bots, I'm playing around with the idea of a chess approach to the levels by mapping out the entire map in player-sized positions. I'm going to refer back to your essay as I run into issues.

The idea of a grid has a lot of potential. There are quite a few obstacle avoidance algorithms that are based on grids, so it should be possible.

It's good to see that someone is taking the essay as an extra source of information (as intended), and not as a complete reference. It is supposed to provide you with interesting options, stimulating thougths, and also reveal the common problems.

Anyway, good luck with your project, and the feedback from your experiments is always welcome here!

interesting but lack of proposal at the end as the conclusion talk about neural nets to find way-points.

What clearly that implies ? how NN can be fixed to do that?