Sensor feedback modification methods that are suitable for the short cycle pattern generation of humanoid walking

A sensor feedback framework that realizes robust walking of humanoids are discussed in the present paper. We have proposed an online walking control system that generates a dynamically stable motion pattern in short cycles, such as 40 [ms]. The system is capable of reflecting the actual motion status modified by sensor feedback to the pattern generation of the next period so that the long-term stability of the walking is maintained using the dynamic model. We herein propose three categories of adaptation as a framework for realizing robust walking via the short-cycle generation system: 1) absorption of the error of the dynamic model, 2) reactive adaptation to the disturbance, and 3) adjustment of the parameters that are used to generate the walking pattern. Feedback methods for each category are discussed and validated on the full-size humanoid HRP-2.