KAIST Uses AI to Create 3D-Printable Soft Robotic Hand Capable of Delicate Grips

Traditional development of soft robotics materials often relied on trial-and-error methods, which are inefficient. A research team led by KAIST changed this status quo by collecting experimental data on various liquid formulations, including material stretchability, hardness, light response, and flow capabilities, to build a comprehensive dataset. This dataset included not only materials that were easy to print but also high-viscosity formulations that were difficult to process.

Based on this dataset, the team trained a machine learning model to identify the relationship between material formulations and their physical properties. The model then screened for a formulation expected to simultaneously meet the requirements for DLP printing and high extensibility. This approach significantly reduced the number of experimental trials needed to develop new 3D-printable materials, allowing scientists to use the model to filter promising candidates before laboratory verification.

Digital Light Processing (DLP) 3D printing technology uses light to cure liquid materials into solid structures, enabling the rapid production of complex shapes. However, the technology faces a long-standing trade-off challenge: components that make materials more durable and stretchable usually increase viscosity, causing them to fail to flow properly during printing; conversely, thinning the liquid to improve printability reduces the material's strength and extensibility.

The research team noted that the newly developed elastomeric material successfully overcame this limitation. Through AI-optimized formulations, the material achieved extremely high mechanical performance while maintaining fluidity suitable for DLP printing. According to researchers, this material can be stretched to more than six times its original length and is resistant to tearing during stretching, thereby reconciling manufacturing feasibility with functional requirements.

To verify the practical performance of the material, the research team used it to print soft actuators. These actuators expand and bend when inflated, simulating the movement of human fingers. The team combined multiple such actuators into a soft robotic hand and tested it under contact-intensive conditions.

Demonstration results showed that the robotic hand could adapt to objects of different shapes and stiffnesses. It successfully grasped fragile eggs, smooth glass bottles, porous egg cartons, and small computer mice. Additionally, the robotic hand could stably lift a water bottle weighing up to one kilogram. This series of actions demonstrated the reliability of the material in handling deformation and bearing loads.

As demand grows for materials that can form complex structures while remaining soft and flexible, this research holds broad application prospects. Such materials are crucial for technologies requiring physical interaction with humans or conforming to irregular surfaces, including soft robotic systems, wearable devices, and patient-customized medical equipment.

Professor Seungchul Lee, the lead researcher, stated that the significance of this study lies in demonstrating that combining experimental data with artificial intelligence can efficiently identify optimal material combinations that were previously difficult to find. This method is expected to accelerate the development of 3D-printable materials with desired properties, serving fields such as soft robotics, wearable devices, and customized healthcare.

This work was supported by the Ministry of Trade, Industry and Energy through its Mechanical and Equipment Industry Technology Development Program and funded by the Ministry of Science and ICT through two research projects. The research findings have been published in the academic journal Nature Communications.

The research team emphasized that the main takeaway from this study is not just the specific material formulation, but the methodology for finding materials. The same machine learning approach can be applied to search for other printable materials with different combinations of mechanical and manufacturing properties, providing a general framework for future materials science exploration.