I cannot remember a time where I was not interested in how people thought. Or more specifically how their brain’s worked. How do we think. Over the years, I’ve studied cognition as a hobby at a macro level with surveys of topics like – “transactional analysis”. It didn’t take long for me to recognize that these were really patterns that sat on top of more complex systems with multiple levels. Naturally this hierarchy leads to neuroscience – specifically the neuroscience of cognition. While I am still deeply curious (after decades), I find that I need a platform upon which to experiment with learning algorithms. Quite accidentally I stumbled upon a really interesting article describe clusters of neurons that independently (endogenously) produce rhythmic output to drive locomotive musculature – I felt inspired to do some experimentation. These structures seem old – they exist in all animals – from simple invertebrates to complex mammals (including humans). They exist in the legless to the multi-legged. The system is not entirely straightforward – as exemplified by concept of the half center oscillator. These are neuron pairs that only exhibit rhythmic output when paired. The ideas that I was most interested in was this passage.
Experiments in which central nervous systems are progressively isolated suggest that, although sensory feedback and the mechanical properties of the musculoskeletal system contribute to this coordination, it also arises in part from central mechanisms. Experiments in limbed vertebrates have shown that individual limbs can produce stepping movements, and experiments in lamprey and leech have shown that a few or even individual segments can produce a basic swimming motor pattern. These data have been interpreted as evidence that each limb, and each or at most a few segments, have their own CPG (a unit oscillator), and that central coordination among limbs and segments arises from coordinating connections among these oscillators.
Borrowing for the electronic world
This concept is the most compelling because one can infer that a loosely couple hierarchy of function exists connecting these systems and the central nervous system. This implies that the process for learning to walk is in turn a layered process for any organism. My hope is that this is model-able in an artifical system using known learning algorithms. Though there are a number of exciting bipedal robots, the hexapods and quadrapods have caught me eye of late – especially after I saw the amazing and eerie BigDog video! The entire video is amazing – but the footage starting at 3:00 is really amazing.
So it is time to start learning about hexapods.