Answers are grounded in this paper and cite their sources.
The coevolution of morphology and behaviour in 3D space has garnered much interest in the field of embodied intelligence. Although recent studies have highlighted the considerable benefits of geometric symmetry for tasks like learning to locomote, navigate and explore in 3D environments, its role within coevolution setup remains unexplored. Existing benchmarks encounter several key issues: (1) the task lacks consideration for spatial geometric information and (2) the method lacks geometric symmetry to deal with the complexities in 3D environments. In this work, a novel setup, named subequivariant morphology‐behaviour coevolution in 3D environments (3DS‐MB), is proposed, which leverages equivariance under the symmetry group to enhance the efficiency and adaptability of morphology‐behaviour co‐evolution. To be specific, EquiEvo is proposed, which injects geometric symmetry, that is, subequivariance, to construct dynamic learnable local reference frames, enabling the joint policy to generalise to diverse task spatial structures, thereby improving coevolution efficiency. Then, EquiEvo is evaluated on the proposed environments, where the method consistently and significantly outperforms existing methods in tasks such as locomotion, navigation and adversarial scenarios, with a significant increase in its learning efficiency. Extensive experiments underscore the importance of subequivariance for the coevolution of morphology and behaviour, effective morphology‐task mapping and robust morphology‐behaviour mapping.
No takes yet. Share an insight, caveat, or question.
Sun et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: