Adaptive Behavior of a Biped Robot Using Dynamic Movement Primitives

摘要

Over the past few years, several studies have suggested that adaptive behavior of humanoid robots can arise based on phase resetting embedded in pattern generators. In this paper, we propose a movement control approach that provides adaptive behavior by combining the modulation of dynamic movement primitives (DMP) and interlimb coordination with coupled phase oscillators. Dynamic movement primitives (DMP) represent a powerful tool for motion planning based on demonstration examples. This approach is currently used as a compact policy representation well-suited for robot learning. The main goal is to demonstrate and evaluate the role of phase resetting based on foot-contact information in order to increase the tolerance to external perturbations. In particular, we study the problem of optimal phase shift in a control system influenced by delays in both sensory information and motor actions. The study is performed using the V-REP simulator, including the adaptation of the humanoid robot's gait pattern to irregularities on the ground surface.