Here are 2 other pics
They show how the collision detection works
The lateral and front traces lengths are function of the bot's velocity. For example, if the bot is strafing left, it will care more about obstacles on its left and little for those on its right, hence the left traces will be fired further than the right traces which will be clamped to the minimum.
here it's the same bot a few seconds later, but seen from the top. Perhaps it's more explicit.
You can notably notice here how the bounding boxes (the boxes used by the engine for collision checking) of any entity in Half-Life are always axial, i.e, parallel to the map axises, no matter the direction the entity is facing. The bot's bounding box is the big green parallelogram.
The three traces per side return different values which are interpreted as: wall (blocking), low wall (jump), low ceiling (duck), fall.
The bot fills then an integer bitmap describing the quality of its surroundings for each of these three sides.
On the second pic for example, you can see in the [body] section of the AI console that the bot senses a low wall on its left, and a low wall in front of it. It can then decide to jump over it.