Abstract A tensegrity-based robot is a locomotive robot that operates on the principle of tensegrity, allowing it to change its shape by adjusting its internal prestress. Tensegrity-based robots can be categorized into different types based on their shape, with the spherical tensegrity-based robot garnering the most attention. However, existing designs for spherical tensegrity-based robots tend to be relatively simple and lack standardized criteria for evaluating their performance. This paper proposes an optimization approach using the force density method to design new spherical regular tensegrity configurations. This is achieved by parameterizing the topology and configuration of the structure, taking into account structural symmetry and the even distribution of internal forces. The proposed approach generates not only classical tensegrities but also novel configurations suitable for locomotive robots. To preliminarily evaluate the suitability of classical tensegrities and novel tensegrities to be used as a rolling robot, a set of performance indexes, including inner space, compactability, prestress evenness, gait repeatability, tilt stability ratio, stride length, and path efficiency, are proposed. The proposed indexes can be quickly determined based on the geometry of the tensegrity and thus are useful in the conceptual selection of the spherical tensegrities for rolling robots. They are used to evaluate a set of six spherical tensegrities. Numerical simulations are carried out to verify the feasibility of geometry-based approximating the gait-dependent indexes. Through the evaluation, a novel spherical tensegrity consisting of 15 struts and 60 tendons is identified as a promising candidate for rolling robots.