the hardware, as usually, is abstractable into drivers.
Take for example hand movement - you may have the model of the robot's hand including actuators' input/output abstractions, length/weight of the arms, etc... It is common for all the robots with arms and very mathematically-mechanically complicated task to calculate the dynamic of the hand from current position, at current speed to another position with another speed, including recoil on the rest of the body, ... Mathematically it is a complicated smooth manifold in the high dimension space and various analyses and optimizations on such objects have been a very fruitful source of many Ph.D.s and articles :) Though on practice any "good" suboptimal solution would do. The 3D orientation and environment sensing is also abstractable - here again the calculation and effective algorithms (which are really hardware independent - for example stereo analysis is dependent basically on the resolution and the speed of the input sensors and teh CPU power and effective algorithm - sounds familiar?) is the heavy part. The [speech] command processing - has been also abstracted into input devices and processing core algorithms.
Take for example hand movement - you may have the model of the robot's hand including actuators' input/output abstractions, length/weight of the arms, etc... It is common for all the robots with arms and very mathematically-mechanically complicated task to calculate the dynamic of the hand from current position, at current speed to another position with another speed, including recoil on the rest of the body, ... Mathematically it is a complicated smooth manifold in the high dimension space and various analyses and optimizations on such objects have been a very fruitful source of many Ph.D.s and articles :) Though on practice any "good" suboptimal solution would do. The 3D orientation and environment sensing is also abstractable - here again the calculation and effective algorithms (which are really hardware independent - for example stereo analysis is dependent basically on the resolution and the speed of the input sensors and teh CPU power and effective algorithm - sounds familiar?) is the heavy part. The [speech] command processing - has been also abstracted into input devices and processing core algorithms.