Goal-Directed CPG-Based Control for Tensegrity Spines with Many Degrees of Freedom Traversing Irregular Terrain

摘要

Abstract To further the ability of robots to achieve goals in environments with irregular terrain, we have developed a series of tensegrity spines as an abstraction of the many degrees of freedom (DOF) compliant spines seen in nature, with full six DOF between vertebrae (constrained by a tensile network). This work provides insight into control strategies for such many DOF and compliant systems, which lack the rigidly connected segments needed by traditional control. Our Central Pattern Generator (CPG)-based controller receives both proprioceptive feedback and goal-directed input. We utilize artificial neural networks to process both the feedback and the input, and only use feedback available to our analogous robotic hardware. This approach seeks to maximize the low-level competence of the control system, by combining local reflexes with structural compliance. This is, to our knowledge, the first example of a robot controlled by CPGs that is simultaneously capable of goal-directed behavior and locomotion on irregular terrain. In addition, this is the first goal-directed controller for a tensegrity robot that can transition between different terrains.