The BBC technology news has today reported on a small robot called Runbot, which has been designed to walk in a way mechanically similar to humans - in contrast to robots such as Asimo, which has each of its movements meticulously calculated and executed using precise control algorithms, leading to its unnatural gait. Runbot takes advantage of morphology and environmental characteristics (such as gravity) in much the same way as Passive Dynamic Walkers (PDW's - such as the Cornell walker), although it moves further by incorporating learning and motor mechanisms to enable walking over slopes (both up and down) and flat surfaces, which PDW's can not. The control architecture is not centralised but distributed, using nested artificial neural networks, and the ability to learn from previous experience. This enables 'normal' walking behaviour to be carried out relatively automatically (as with the PDW's), but when the circumstances change (for instance when an incline is encountered), higher level processes use the previously learned information to mainting balance and forward momentum. In this way, speeds of up to three leg lengths per second are achievable, which is only slightly slower than the fastest walking human.
UPDATE 14/07: Madam Fathom has a much more detailed overview of the Runbot here, describing central pattern generators, and the hierarchical organisation. Well worth a look.