Development of a Platform to Evaluate Principles of Bipedal Locomotion Using Dynamical Movement Primitives

The control of bipedal locomotion is often considered to arise from the optimization of variables related to energetic cost and stability. However, developing model-based predictions of optimal strategies can be challenging due to the high-dimensionality of the control space and challenges associated with optimizing potentially conflicting objectives. Here, we present a framework for simulating bipedal gait which can ultimately be used predictive simulations of optimal gait patterns. We modeled a human-like biped and a treadmill in Matlab Simscape and used Dynamical Movement Primitives (DMP) to generate control joint-level controllers from demonstrations of human walking. DMPs facilitate the optimization of gait patterns through a small number of free parameters, such as the amplitude and timing of the patterns. We also implemented a simple feedback controller that variated the amplitude of the DMPs to stabilize the biped based on the deviation from a reference point on the treadmill. Optimizing of this controller allowed us to generate a human-like gait and ultimately contributed to the development of a platform with which we can explore optimization principles during locomotion.