passive dynamic walkers

Generating patterns

I’ve only fairly recently come across the videos of the results of the OpenSim dynamic walker competition on the OpenSim YouTube channel. The OpenSim team made the basic models available and challenged people to see how far they could make the models walk. Although you are too late to participate in the competition all the material is still available on their web-site to allow you to prepare a late entry. There are really two challenges, one is to design the model and the other is to work out exactly how to “push” it to get it started. The three best results are linked to in the video below.

Dynamic walkers are simple mechanical structures with a small number of links which, when pushed in a particular way to get started, will walk down a slope (they have to walk down a slope because they lose some energy whenever the foot collides with the ground). They were pioneered by Tod McGeer over twenty years ago and this video shows him explaining the principles. The field has grown quite markedly since and there is an annual Dynamic Walking conference (the last was in Zurich).

It seems obvious that we can learn  about the mechanics of walking using these models (although it is also important not to over emphasize the similarities – all passive walkers used a locked knee during stance, for example, whereas most of us walk in a little knee flexion which is not inherently stable and requires muscle activity). It stuck me looking at these videos that walkers may also be able to tell us something about neurology – principally because they don’t have any. These walkers are simple assemblies of passive mechanical components and without any control mechanisms at all.

What interests me is that a lot of people in the field of motor control seem to assume that if a complex cyclic pattern is observed then there must be something in the central nervous system generating it. Perhaps the most commonly cited example is the decerebrate cat. If you are careful, you can take the brain (cerebrum) out of a cat, place it on a treadmill and it will walk, and may trot and gallop as the speed increases (you can see a short video here – but you don’t have to if you object to this sort of animal experimentation). This is not the only evidence that is cited in support of the assumption that walking is driven by central pattern generators in the spinal cord (see the first two sections of this this paper for example) but it is the most graphic.

The passive dynamic walkers, however, do not have any control system for generating patterned movement and still exhibit complex cyclic patterns of movement.  It seems clear to me that whilst the basic gait pattern might be a consequence of pattern generating neurons somewhere in the central nervous system it doesn’t necessarily have to be. I’m not an expert in the field but I’d guess that the most likely explanation is that there is an interaction between the basic mechanical behaviour of the system and the neural control which is rarely discussed. This is supported by recent work from Gottschall and Nichols which shows that, however the decerebrate cat is generating cyclic movement, the walking pattern is influenced by head position, indicating that it is modulated by either vestibular or proprioceptive input.  As Art Kuo pointed out some years ago control systems incorporating proprioceptive feedback are likely to be more stable than those driven purely by pattern generators (feed-forward control).