Using generative AI to help robots jump higher and land safely
Generative Artificial Intelligence (GenAI) models, widely known for their capabilities in design brainstorming with tools like OpenAI’s DALL-E, are now making significant inroads into the practical development of robotics. Moving beyond conceptual images or blueprints, these advanced AI systems are actively contributing to the creation and optimization of functional robots, transitioning from theoretical designs to high-performing physical machines.
A pioneering new approach from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) exemplifies the transformative power of generative AI in enhancing robotic designs. Researchers have devised a methodology that allows users to provide a 3D model of a robot and designate specific components for AI modification. The GenAI system then iteratively brainstorms and rigorously tests optimal shapes for these areas within a simulated environment. Once an ideal design is pinpointed, it can be seamlessly saved and fabricated using a 3D printer, effectively streamlining the design-to-production pipeline by eliminating the need for additional manual adjustments.
This innovative, AI-driven optimization has yielded impressive results. Through this process, researchers successfully engineered a robot capable of an average leap of approximately 2 feet, representing a remarkable 41 percent improvement over a similar machine designed solely by human engineers. While visually almost identical—both constructed from polylactic acid and transforming from a flat state into a diamond shape when a motor activates a connecting cord—the crucial differentiation lies in the AI’s subtle yet profound design alterations.
A closer inspection reveals that the AI-generated linkages, integral to the robot’s jumping mechanism, possess a distinctive curved shape, reminiscent of a thick drumstick. In stark contrast, the human-designed counterpart featured straight, rectangular connecting parts. This seemingly minor design change, conceived by the AI, is critical: it enables the robot to store significantly more energy before each jump, a counter-intuitive solution that often eludes traditional human design approaches that might prioritize thinness for lightness.
Byungchul Kim, co-lead author and CSAIL postdoc, emphasized the unique problem-solving capabilities of diffusion models. “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,” Kim explained. “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 AI’s prowess extended beyond mere jumping ability. The research team also tasked the system with designing an optimized foot for enhanced landing stability. Through a similar iterative optimization process, the best-performing design was selected and integrated into the robot. This AI-designed machine demonstrated an astounding 84 percent improvement in landing stability, drastically reducing instances of instability compared to its baseline version. This dual enhancement in both jumping height and safe landing underscores the profound potential of generative AI in tackling complex engineering challenges.
The versatility demonstrated by this diffusion model suggests extensive applications for improving various types of machinery. Industries involved in manufacturing or household robotics, for instance, could leverage similar AI-driven approaches to refine their prototypes, significantly reducing the considerable time and resources typically allocated to manual iteration and design adjustments. The researchers anticipate expanding into even more flexible design goals in the future.
Tsun-Hsuan “Johnson” Wang, co-lead author and MIT CSAIL PhD student, envisions a future where “natural language [can] guide a diffusion model to draft a robot that can pick up a mug, or operate an electric drill.” Further advancements could involve incorporating lighter materials to achieve even higher jumps, generating articulation for improved part connections, and integrating additional motors for precise directional control and enhanced landing stability.
This groundbreaking work, which received support from 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, was presented at the prestigious 2025 International Conference on Robotics and Automation. This presentation marks a significant milestone in the rapidly evolving synergy between artificial intelligence and robotics.
