Artificial intelligence news.
Straight from the original announcements.

MIT AI copilot lets engineers guide structural designs as they evolve

November 13, 2026, 06:00By the AI News Bites Editorial TeamSource: MIT News
MIT AI copilot lets engineers guide structural designs as they evolve
Image: MIT News

MIT researchers have developed an AI copilot that asks engineers for feedback while an algorithm generates designs for structures such as buildings and bridges. The approach aims to make computer-generated designs easier to manufacture and more trustworthy without substantially extending the design process.

Key points
  • MIT researchers created an AI copilot for topology optimization.
  • The system presents design regions to engineers for feedback as structures evolve.
  • The approach added 15 seconds to total design time in the reported comparison.
  • Demonstrations showed a 39 percent improvement in manufacturability and linear buckling load.
  • Users can currently make identified regions thicker or thinner.

Human feedback during the design process

Topology optimization uses formal optimization techniques to fill a defined design space while pursuing goals such as structural performance. Although the method has produced parts for ultralight aircraft, high-bandwidth antennas and F1 racing cars, it can require substantial computation and time.

With conventional approaches, engineers may see the completed design only after the algorithm finishes. If the result contains impractical features, such as eight or more beam connections at one point, they may have to restart the process. MIT says this can make the technology harder to use for buildings and bridges, where engineers need control over how a structure develops.

The researchers' system presents selected regions to a human designer while the structure is being generated. The designer can currently make structures in those regions thicker or thinner. The system uses machine learning to identify areas where a person would most likely want to intervene, rather than requiring feedback across the entire design.

Machine learning identifies likely intervention points

To train the machine-learning model, the team created a synthetic dataset containing probable human feedback on thousands of designs. The feedback addressed factors including aesthetics, performance criteria and manufacturability.

In an open-access Communications Engineering paper, the researchers reported that the human-in-the-loop approach produced structures that were easier to manufacture and had higher buckling loads. MIT News says the system added only 15 seconds to the total design time compared with all-algorithmic approaches.

In demonstrations, the researchers reported a 39 percent improvement in manufacturability and linear buckling load, while the total design time increased only modestly. The researchers said the system brings together algorithmic optimization with the experience and judgment of engineers.

“The idea is to tap into the brains of the designers and get them to infuse their domain knowledge as the design evolves,” senior author Josephine Carstensen, an MIT associate professor of civil and environmental engineering, said in the MIT News report.

A route toward more trusted structural designs

Lead author Dat Ha said engineers can be skeptical of computer programs that produce unexpected designs, describing topology optimization's “black box nature” as a barrier to wider industry adoption. Carstensen said the lack of control and ownership has also limited its use.

The researchers do not present the copilot as a replacement for algorithm-only design. Carstensen said both approaches have a role, depending on the available computational resources and the need for human input. In some settings, engineers must make the final decisions before a design advances.

The current system has limited modification options, but the team hopes to add more in future versions. According to MIT, the researchers believe this could make topology optimization more accessible to design engineers, researchers and students, while helping produce structures that people will accept and build.

Related news

MIT’s FLEX Helps Robots Infer More Than Human Commands Say
ResearchNovember 9, 2026, 06:00

MIT’s FLEX Helps Robots Infer More Than Human Commands Say

MIT and New York University researchers have developed FLEX, an AI-assisted learning platform that helps robots infer broader meaning from conversational commands. The…

MIT researchers build one AI model for sand, fluids and proteins
ResearchNovember 9, 2026, 06:00

MIT researchers build one AI model for sand, fluids and proteins

Researchers at MIT, the University of British Columbia and other institutions have developed WorldParticle, a single deep-learning system designed to simulate a wide…

MIT system lets robots find and fix wrong object labels in memory
ResearchNovember 5, 2026, 06:00

MIT system lets robots find and fix wrong object labels in memory

MIT researchers have developed UQ-DAAAM, a memory system that identifies unreliable object descriptions stored by a robot and corrects them while the robot moves.…