Perturbation theory to plan dynamic locomotion in very rough terrains

Abstract

The control of tendon-driven robots using techniques from traditional robotics remains a very challenging task that has been so far only successfully achieved for small-scale setups comprising exclusively revolute joints [1, 2]. Hence, we propose a fundamentally different approach. Instead of deriving an analytical robot model using either the Newton-Euler or Lagrangian formulation we suggest to employ physics-based simulation engines to simulate the peculiar dynamics of this emerging class of robots and to use the simulated robot model as an internal model for robot control [3]. In this paper, we present the reverse-engineered derivation of a detailed physics-based model of an anthropomimetic robot implemented on CALIPER [4], a simulation framework developed within the EU-funded project ECCEROBOT [5]. The model comprises an accurate model of the skeleton derived from laser scan data, as well as of artificial ligaments and muscles. The individual sub-models are validated separately against measurements and the successful integration of all sub-models is demonstrated by executing a limb movement which requires the parallel control of multiple muscles.