COSMIC robot co-design framework optimizes structure, material, and control simultaneously

Most robotic systems today are designed in silos—structure, materials, and control are optimized separately, contrasting sharply with nature's co-evolution. A new paper from Qinsong Guo and Liwei Wang introduces COSMIC (Concurrent Optimization of Structure, Material, and Integrated Control), a gradient-based framework that simultaneously optimizes all three entities for truss-lattice robots. The key innovation is embedding mixed-type topological and material variables into a continuous design space, then integrating a neural network controller within a differentiable simulator. This allows automatic differentiation to compute gradients efficiently, capturing complex interactions between design decisions.

Case studies reveal that COSMIC consistently discovers diverse locomotion strategies that outperform baselines from separated design approaches. The framework also incorporates constrained optimization to navigate highly non-convex design landscapes. Beyond performance gains, it extracts design insights showing the individual and collective contributions of structure, material, and control—something previously difficult to isolate. COSMIC is flexible enough to handle different functional requirements and boundary conditions, paving the way for fully autonomous co-design of reconfigurable, locomoting robots. The work represents a computational foundation for systems that can evolve their own form and function, much like natural organisms.