AI-Enhanced Robots Leap Higher and Land Safely

Diffusion models, such as OpenAI’s DALL-E, are increasingly useful for brainstorming new designs. These models can generate images, videos, and blueprints, offering ideas that humans might not consider. But did you know that generative artificial intelligence (GenAI) is also making significant strides in creating functional robots?

Recent diffusion-based approaches have generated structures and control systems from scratch. These models can create new designs, simulate their performance, and refine them before fabrication. A new approach from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) leverages this generative know-how to enhance human-designed robots.

Users can draft a 3D model of a robot and specify which parts they want the diffusion model to modify, providing dimensions beforehand. The GenAI model then brainstorms the optimal shape for these areas, tests its ideas in simulation, and saves the best design for fabrication with a 3D printer.

The researchers used this method to create a robot that jumps an average of roughly 2 feet, or 41 percent higher than a similar machine they designed manually. Both robots are made of polylactic acid and appear flat but spring into a diamond shape when a motor pulls on a cord attached to them. The key difference lies in the AI-generated linkages, which are curved and resemble thick drumsticks, while the standard robot’s connecting parts are straight and rectangular.

The researchers began by sampling 500 potential designs using an initial embedding vector, selecting the top 12 based on simulation performance. This process was repeated five times, progressively guiding the AI model to generate better designs. The resulting design, which resembled a blob, was scaled to fit the 3D model and fabricated. The AI-designed robot indeed jumped higher.

Byungchul Kim, a CSAIL postdoc and co-lead author, explains, 'We wanted to make our machine jump higher, so we figured we could just make the links connecting its parts as thin as possible to make them light. However, such a thin structure can easily break if we just use 3D printed material. Our diffusion model came up with a better idea by suggesting a unique shape that allowed the robot to store more energy before it jumped, without making the links too thin. This creativity helped us learn about the machine’s underlying physics.'

The team then focused on optimizing the robot’s foot to ensure it landed safely. They repeated the optimization process, eventually choosing the best-performing design. The AI-designed robot fell far less often, showing an 84 percent improvement in landing stability.

The diffusion model’s ability to enhance a robot’s jumping and landing skills suggests it could be valuable in other areas. For example, a company manufacturing or household robots could use a similar approach to improve prototypes, saving engineers time.

To create a robot that could jump high and land stably, the researchers balanced both goals by representing them as numerical data and training their system to find a sweet spot between both embedding vectors. This AI-assisted robot outperformed its human-designed counterpart, but the researchers believe future versions could jump even higher with lighter materials.

Tsun-Hsuan ‘Johnson’ Wang, a CSAIL PhD student and co-lead author, says the project is a starting point for new robotics designs. 'Imagine using natural language to guide a diffusion model to draft a robot that can pick up a mug, or operate an electric drill.' Kim adds that a diffusion model could also help generate articulation and improve how parts connect, potentially enhancing the robot’s jumping height. The team is exploring adding more motors to control the direction the machine jumps and improve landing stability.

The researchers’ work was supported by the National Science Foundation’s Emerging Frontiers in Research and Innovation program, the Singapore-MIT Alliance for Research and Technology’s Mens, Manus and Machina program, and the Gwangju Institute of Science and Technology (GIST)-CSAIL Collaboration. They presented their findings at the 2025 International Conference on Robotics and Automation.