Motion Planning for Continuum Robots: A Learning from Demonstration Approach

Abstract

Continuum robots have been recently used in the inspection of tight and confined spaces. State-of-the-art motion planning algorithms that are developed for rigid robots could be inadequate when applied to redundant and complainant continuum robots. In this research, a Demonstration-Guided Motion Planning (DGMP) framework is proposed to let continuum robots imitate a set of given demonstrations to plan and execute point-to-point spatial motions. A flexible interface is incorporated to allow humans to intuitively demonstrate motions for the robot via teleoperation. The Dynamic Movement Primitives (DMP) framework is adopted to learn, reproduce and generalize the given demonstrations while avoiding static or moving obstacles existing in the environment. Meanwhile, a Model Reference Adaptive Controller (MRAC) is proposed to ensure the robot robustness against perturbation while tracking the generated motions from the DGMP. The developed approach is evaluated over a simulated model of a two-section continuum robot. The obtained results show evidence that the proposed DGMP is effective in generating and tracking spatial motions for continuum robots. This could encourage a further investigation towards planning complex motions in the future for redundant continuum robots.