MIT CSAIL researchers have successfully utilized generative AI to enhance the design of a jumping robot, resulting in a machine that outperforms its human-designed counterpart. The process…
MIT CSAIL researchers have successfully utilized generative AI to enhance the design of a jumping robot, resulting in a machine that outperforms its human-designed counterpart. The process involves creating a 3D model of a robot and specifying areas for the AI to modify. The generative AI then brainstorms and tests optimal shapes through simulation.
The researchers used this approach to create a robot that leaps up an average of 2 feet, or 41 percent higher than a similar machine they created on their own. The AI model was trained by sampling 500 potential designs and selecting the top performers for optimization. The resulting design featured curved, drumstick-shaped linkages, unlike the straight, rectangular parts of the standard robot.
The AI also designed an optimized foot, leading to an 84 percent improvement in landing stability. This demonstrates the potential of generative AI to find unconventional solutions and improve robot performance by storing more energy before jumping and ensuring safer landings. The researchers' approach balances jumping height and landing success rate by representing both as numerical data and training the system to find an optimal 3D structure.
The team is exploring future iterations with lighter materials to achieve even greater heights. They also envision using natural language to guide the AI in designing robots for various tasks, such as picking up objects or operating tools. This research highlights the potential of generative AI in robotics design, offering a way to improve existing designs and potentially saving time for engineers.
The researchers aim to expand the application of generative AI in robotics to create machines with enhanced capabilities. The project was supported by various organizations, and the findings were presented at the 2025 International Conference on Robotics and Automation.