AI News

Using generative AI to help robots jump higher and land safely | MIT News | Massachusetts Institute of Technology

MIT CSAIL researchers have successfully employed generative AI to enhance the design of robots, resulting in a machine that outperforms its human-designed counterpart in both jumping height…

Using generative AI to help robots jump higher and land safely | MIT News | Massachusetts Institute of Technology

Jul 6, 2025

Using generative AI to help robots jump higher and land safely | MIT News | Massachusetts Institute of Technology

MIT CSAIL researchers have successfully employed generative AI to enhance the design of robots, resulting in a machine that outperforms its human-designed counterpart in both jumping height…

MIT CSAIL researchers have successfully employed generative AI to enhance the design of robots, resulting in a machine that outperforms its human-designed counterpart in both jumping height and landing stability. The team utilized a diffusion model, similar to those used for image generation, to refine the structural components of a jumping robot.

By providing a 3D model and specifying areas for modification, the AI generated and tested various designs in simulation, ultimately optimizing the robot's linkages for improved performance. The AI-designed robot showcased a 41% increase in jump height compared to the human-designed version.

The generative AI identified unconventional solutions, such as curved, drumstick-shaped linkages, which allowed the robot to store more energy before jumping without compromising structural integrity. Furthermore, the AI was tasked with designing an optimized foot, leading to an 84% improvement in landing stability.

The researchers iteratively refined the robot's design by sampling numerous potential configurations and optimizing based on simulation performance. This process, guided by numerical representations of jumping height and landing success, allowed the AI to strike a balance between these competing goals.

The team believes this approach could be applied to other robotic designs, saving engineers time and improving performance in various applications. Looking ahead, the researchers envision expanding the capabilities of generative AI in robotics, potentially using natural language prompts to design robots for complex tasks like manipulating objects or operating tools.

They also plan to explore the use of lighter materials and additional motors to further enhance the robot's jumping ability and maneuverability. This work highlights the potential of GenAI in creating innovative and efficient robotic designs.