MIT AI Copilot Helps Engineers Design Better Structures

MIT researchers have developed an AI copilot for structural design that inserts human feedback into topology optimization, an algorithmic method used to generate high-performance engineering structures. The system aims to make computationally intensive design workflows more collaborative, manufacturable and trustworthy, addressing a key barrier to wider industrial adoption. Topology optimization iteratively allocates material within a given design space to meet objectives such as strength, weight or performance. The technique has produced optimized components for ultralight aircraft, high-bandwidth antennas and Formula 1 racing cars, and has gained traction in aerospace and automotive engineering. However, the process can require extensive computation, and engineers typically see the result only after completion. If the output contains impractical features, difficult geometry or unexpected beam connections, users may need to restart the process. That black-box workflow has limited confidence in the method, especially for large civil structures such as buildings or bridges. The new MIT approach treats topology optimization as a collaborative design process. While the algorithm generates a structure, the system identifies regions where human designers are likely to want changes, such as those affecting manufacturability, performance or aesthetics. Machine learning guides the process by prioritizing the parts of the design most likely to require human judgment. To train the model, the researchers created a synthetic dataset containing probable human feedback across thousands of designs. According to the team, the method increases engineer agency without significantly slowing computation. In demonstrations described in an open-access Communications Engineering paper, the human-in-the-loop approach improved manufacturability and linear buckling load by 39 percent, while adding only about 15 seconds to the total design time compared with fully automated workflows. Current versions allow users to modify identified regions by making structures thicker or thinner, with broader editing options planned for future work. The project was led by Dat Ha PhD ’26, a recent MIT civil and environmental engineering doctoral student now working as a structural designer at LeMessurier. Co-authors include Ferdous Alam, Markus Buehler, Faez Ahmed and Josephine Carstensen. The researchers argue that the method does not replace fully automated AI design, but complements it by allowing engineers to inject domain knowledge during generation rather than after the fact. By involving humans earlier and more directly, the team says the system can improve trust, ease industrial implementation and help move higher-performing designs from simulation into real-world structures.