Design optimization and integration of deployable cellular mechanisms using deep reinforcement learning

With the rapid advancement of aerospace technology, deployable mechanisms are playing an increasingly important role. In this paper, a design method for deployable cellular mechanisms based on deep reinforcement learning is proposed and then applied to construct a class of asymmetric multi-loop plane-symmetric Bricard cellular mechanisms. First, the advantages of quadrilateral geometric units in constructing large-scale deployable mechanisms are analyzed, comparing the flexibility of symmetric and asymmetric configurations in building large deployable mechanisms. Subsequently, the Deep Deterministic Policy Gradient (DDPG) algorithm is introduced. A virtual simulation environment eliminates the need for closed-loop constraint equations, enabling adaptive optimization of key structural parameters to get a cellular mechanism with deployment consistency. Moreover, a tessellation strategy of alternating convex-concave 3R chains is proposed, demonstrating its potential for large-scale assembly in ring, linear and planar configurations. Furthermore, a formula for calculating the folding ratio is provided to quantitatively evaluate the mechanism's deployment performance. And prototype experiments validate the effectiveness of the designed mechanism. The reinforcement learning-based mechanism design method offers a framework for lightweight and large-scale design of deployable mechanisms in aerospace applications.