Fig. 4: State and temporal dynamics of the robot during task performance with random high-level commands.

a,b, Sagittal motion of the CoM while following random leg (a) and arm (b) commands, respectively. From the robot snapshots corresponding to the time they’re shown, the partial autonomy of the mid-level stability controllers can be seen, that is, a good performance of the individual levels despite random and fast-changing command inputs. c,d, Leg (c) and arm (d) movements, respectively. Here, the separation of temporal scales during planning can be seen, where the high-level commands are provided at 0.5 Hz and the mid-level commands are realized at 25 Hz. The joint commands are realized at 500 Hz on the joint actuators. In the inset plots of c,d the joint position trajectories evolve similarly as postulated in the equilibrium point hypothesis.