A Human-like Walking for the Compliant Humanoid COMAN based on CoM Trajectory Reconstruction from Kinematic Motion Primitives

Research on humanoid locomotion made significant improvements over the last years. In most cases, though, the gait of state-of-the-art robots is still far from being human-like due to two main reasons. These are, the mechanical incompatibilities between the human and the engineered humanoid platforms, and the lack of clear understanding of the highly complex human walking motion itself. This work attempts to address the latter by using a novel method to construct locomotion trajectories for a humanoid robot based on kinematic Motion Primitives (kMPs) derived from humans locomotion trajectories. The work demonstrates how from a small set of invariant primitives it is possible to reconstruct all the joint trajectories to obtain different gaits, with different speed, and while accomplishing different tasks at the same time. We then used the proposed method to reconstruct a human-like CoM trajectory, and evaluate it on our Compliant huMANoid robot, COMAN. Experimental results are presented to demonstrate the execution of a stable, fast walking, with knee straightening at toe-off to push forward, that strongly resembles a more human-like walking. Furthermore, and taking inspiration from the inherent passive compliance in the joints of the COMAN robot, the energy consumption of the robot was investigated by varying the frequency of the gait generated basing on the human kinematic primitives. The idea was to exploit the natural dynamics of the compliant humanoid platform, and minimize its energy consuption. Experimental results for different stepping frequencies in terms of energy consumption over a fixed time, and to walk for a fixed distance are presented.