My understanding of walking robots (e.g. https://www.youtube.com/watch?v=xJlkBBdyBYI) is that they use a gyroscope to determine the current orientation of the robot, or each joint of the robot. This is because if you just put encoders on each joint, the cumulative error over the entire robot will be too large to maintain stability. Therefore, a gyroscope measures the "real" orientation, and this is used for feedback when the robot is walking.
However, I'm also aware that some walking robots use accelerometers to maintain stability. What would be the benefit of using an accelerometer in this case? Would it be used instead of a gyroscope, or together with a gyroscope?
My guess is that gyroscopes do not measure acceleration directly (unless you were to numerically calculate this based on lots of orientation readings), but accelerometers do measure it directly (and more reliably than this numerical method). Know the acceleration as well as the position then enables to robot to more accurately predict its future position, and hence the feedback loop is more robust. Is this correct, or am I missing the point?