When I was a teenager acting out the customary role as Future Comic Artist (by which I mean copying and memorizing individual poses out of comic books), it was unsurprising that I became obsessed with the idea of anatomy. I felt that "learning all the muscles" was going to magically make me a better artist, though it seemed like a daunting and impossible task. I dutifully copied some diagrams from an anatomy book, and felt a little tinge of pride knowing what a deltoid was, or that the calf muscle was called the gastrocnemius (I skipped over the soleus). However, I had skipped over two huge facets of anatomy: the skeleton, and perspective.
George Bridgman used a brilliant phrase as the title of one of his books on drawing the figure: "The Human Machine". It's a good way to think about the figure. From a sculptural point of view, a good exercise would be to build an actual machine out of metal and wood that mimics the functionality and range of motion of the human skeleton, using parts you find at the hardware store. The goal would be to build something that could move in the same way as a human, but it wouldn't necessarily need to be shaped like one. If we had some space-age synthetic muscle, we could rig it up to actually move. What kind of joint would you need for the shoulder? What would you make a vertebra out of, and how would it connect to its neighbor and how would it move? How would you construct a pelvis so that there were joints in the proper places?
The good news is that we don't actually need to build this machine- it's very helpful just to plan it out mentally. The more accurately and true-to-life we can plan this machine, the better our understanding of the figure will be. The bad news is that as two-dimensional artists we need to be able to draw it. We don't necessarily need to be able to draw the actual parts we'd get from the hardware store- we can use some imaginary parts that have the same types of mechanical movement, and we can use simple masses for the bones.
This inevitably gets back to perspective. It's not enough to understand what a particular joint would be, we also need to be able to draw that joint and the major axes of the bones involved from any angle. Further, once we draw two joints, we also need them to be arranged in a way that would make sense given a single, fixed viewpoint and the ranges of motion for the joints involved. One of the most common errors in figure drawing is drawing the different masses in a way that implies either that either the joints are in the wrong location or moved past their possible range of motion, or that the masses are in fact being viewed from different viewpoints.
Let's take one of the more simple joints in the skeleton- the elbow joint of the humerus and ulna. In other words, the joint that allows you to bend your forearm up to do a bicep curl. If we ignored the radius for now (which allows you rotate your palm), we could build this joint very simply for our machine: just a couple two-by-fours and a regular door hinge. When drawing, we can represent it with a long, thin box for the humerus, a long, thin box for the ulna, and an imaginary cylinder joint between them for the hinge. Assuming the "elbow" isn't bent, things are pretty easy to draw, right? From the sides and front, it's a piece of cake. You could probably even tip it around in space fairly easily as well.
However, once we bend that elbow a bit, things get complicated. From the sides it's not too hard, and using that information we can estimate the front view fairly well (we're fudging the foreshortening). But once we rotate our bent contraption, we get into trouble:
See? Right away we are back at perspective. How would you solve this particular problem? If you have no idea, then how are you going to draw the arm in a similar position? Even if you have the model posing in front of you, it's beneficial to understand that this particular perspective problem is at play. If the pose is short enough, you will need this understanding to get something legible down quickly. If the pose is longer (particularly if there are multiple breaks) you will need this understanding to compensate for shifts in the pose.
This is for one of the simplest joints in the body. It's further complicated by the fact that the ulna does not actually line up with the axis of movement created by the cylinder joint- it's slightly skewed off in what's called the "carrying angle" of the lower arm (named because it allows space to carry objects such as a full bucket without it hitting your thigh). To complicate things even more, the carrying angle varies from person to person, especially between men and women (due to the relatively wider pelvis and narrower shoulders of the female skeleton). And we haven't even thrown in the radius.
A good first step to this kind of understanding is to start planning out your human machine, and understanding how to represent it in 3-dimensional space from a fixed viewpoint. Start simple. Start with what you understand. When you find a gap in your understanding, look for an answer. When you find a discrepancy between your mental machine and the model or a piece of reference, adjust the design for your machine accordingly. Let your machine evolve along with your understanding. As it does, you'll find both drawing from the figure and from imagination much less mystifying.