The Quality of Movement

The human body has 244 joints and 640 muscles, enabling an infinite set of possible movement patterns to reach any point in space. This raises a crucial question: what's the optimal movement strategy?

While efficiency is the obvious driving force behind movement quality, it's a complex optimisation problem when considering a highly redundant system like the human body. Efficient movements must be locally optimal within the constraints of body structure: Joints, Muscles, and Neural pathways, and globally optimal with respect to task completion or organism survival. For example, Olympic javelin thrower Neeraj Chopra may have biomechanics that mean he can throw a ball over a wall more efficiently than me, while a dog might opt to carry it around instead. A chihuahua might abandon the task entirely, expressing its displeasure with frustrated evil grunts.

Thankfully, humans possess a remarkable ability to identify efficient movement patterns given constraints. The body finds ways to complete tasks despite injuries, often so naturally that we take this ability for granted. Consider how someone who sprains their ankle while playing football instantly develops a compensatory limp. This adaptation is not just pain avoidance—in an otherwise neurologically intact body, the limp is adjusted to be the most efficient way to navigate the world with a sprained ankle.

However, optimal movement isn't automatic in all cases. Compensatory movemenets if maintained for long enough can become habitual, and even if the underlying injury heals, the brain may continue to use the compensatory pattern. This is not all too rare and can lead to long term movement dysfunctions. Worse yet, stroke patients often spend years relearning basic movements like walking and grasping, highlighting the complexity of our movement patterns and the neural systems controlling them.

It's no surprise then that identifying efficient movement patterns has a lot of practical benefits. And we can kinda do it already, except it takes a lot of effort. Coaches can identify inefficiencies in an athlete's movement patterns, and help them improve. Physical therapists can identify dysfunctions and help rehabilitate injuries. Numerous tools like motion capture systems and force plates can help quantify movement patterns. The problem is that it's hard to do it well, and it's hard to do it fast.

Human experts are historically good at this, but they are hard to scale. A degree is not expertise, years of experience is. And even then, information is scattered and siloed. AI seems like it could be the answer, but current models are stumped by a huge lack of training data. Data collection is hard, and given the expertise requirement, annotation is even harder. This is felt in a different context, in the best of sports science departments of educational institutions, when teaching human students. Their strategy? to simulate compensatory movements in healthy experts to teach students how to identify them. But, this creates a general mapping of what to look for. And this maybe applicable to ML researchers as well. A dense dataset of simulated human movement patterns could be used to train models that atleast have a fighting chance of identifying compensatory movements irl.