Innovation Design of Multilocomotion Planar Tensegrity Mobile Robot Driven by Machine Learning

Abstract Tensegrity mobile robots, known for robust terrain adaptability and impressive stiffness-to-mass ratio, have become a focal point in research. Particularly, the innovative design of multilocomotion tensegrity robots is challenging due to their complex parameters and flexible components. Achieving the design of a multilocomotion tensegrity mobile robot, which encompasses both topology and dimension, remains a significant challenge. Here, we introduce a design approach combining machine learning technology and incremental model. Utilizing our approach, the redundant design parameters (such as topology, dimension, stiffness of elastic cable, and driving laws under specific terrains) of the tensegrity robot could be acquired without complex mathematical models and intricate iterative design processes. The proposed innovative design approach yields an approximately 12.47-fold increase in efficiency for obtaining the first feasible tensegrity robot configuration compared to enumeration, thereby significantly reducing development time and improving overall design efficiency. Furthermore, our machine learning-driven design approach has demonstrated robust generalization capabilities, facilitating the formulation of driving laws for navigating unknown terrains, thus paving new ways for mobile robots.