Fig. 3: Significance of the stabilizing feedback controller: avoiding falling and improving learning.
From: Exploration-based learning of a stabilizing controller predicts locomotor adaptation
a The default controller provides robust stability to the biped despite noise and environmental changes. Substantially lowering the feedback gains, all by the same factor, reduces the effective adaptation rate and increases gait variability. The sensory noise in these simulations is fixed across these feedback gain conditions and is added to velocity feedback to the feedback controller. Adaptation phases are shaded in blue. b Lowering the feedback gains even further results in falling of the biped upon introducing the split-belt perturbation. Three walking patterns are shown: normal tied-belt walking that has symmetric step lengths, split-belt walking with default feedback gains resulting in stable but asymmetric gait, and split-belt walking with much-reduced feedback gains resulting in falling. In the bottom-most row, the center of mass trajectories for each stance phase are shown relative to the respective stance belt frame for visualization purposes (so that the split-belt trajectories for the different stance phases are with respect to different frames).
