Sparse-code muscle representation for human walking

Research in robotics has recently broadened its traditional focus on industrial applications to include natural, human-like movements. Progress has been made on construction of actuators, like the Pneumatic artificial muscles (PAMs) and the Shadow Air Muscle, that have properties similar to those of human muscles. There is, however, still a need for efficiently representing the movements spanned by human muscles. Given the number of joints and muscles in the human body, there are typically an enormous number of ways to perform any particular action, thus there is a strong motivation to parametrize these possibilities in some helpful way. The objective of this research work is to understand and simulate primitives that lead to various smooth natural movements. In this paper we propose a sparse-code representation for muscle length activations obtained by applying Basis Pursuit on such movements. We employ accurate three-dimensional musculoskeletal models to simulate the lower body movements for walking that are captured from a human subject using PhaseSpace motion capture system, and translated into muscle lengths using the SIMM system. We recreate the same muscle length changes, and show that the signal can be economically encoded in terms of discrete movement elements. Each movement can thus be visualized as a sequence of coefficients for temporally displaced motor primitives.