Skip to Main Content

When robots are built with biological materials, they have the potential to achieve remarkable behaviors typically only seen in nature. For example, unlike traditional actuators, actuators built from muscle tissue can adapt and grow stronger with use. This means that a robot powered by living muscle doesn’t just move—it exercises and gains the ability to adjust to its environment and perform tasks more efficiently over time.

In order to deploy biohybrid robots for specific tasks, researchers must first understand how to design and control robots who can get stronger over time. The Biohybrid and Organic Robotics Group, led by Vickie Webster-Wood, has created a model to support just that. By using reinforcement learning, their approach learns to control a model of a biohybrid robot, even as the muscles get stronger each time they attempt the task.

3D visualization of a dynamic structure represented in a wireframe model, showing two different stages of a simulation in physics.

To test this, a soft, worm-like robot made up of 42 living muscles was tasked with moving towards eight different targets in a simulated environment. To reach each target, the robot needed to learn to coordinate its muscles differently. To understand how muscle exercise would affect the ability of the controller to learn to reach the objects, the team ran simulations with static muscles and muscles that got stronger with use.

“At the start of this experiment, we questioned whether or not the AI agent would be negatively impacted by muscle adaptability,” said Webster-Wood, associate professor of mechanical engineering. “However, we found that having adaptable actuators didn’t hurt learning at all.”

This brings us one step closer to designing and eventually building biohybrid robots that can adapt to the world around them, just like animals do.

Vickie Webster-Wood, Associate Professor, Mechanical Engineering

The team successfully taught the robot how to move toward eight different targets by coordinating its muscle contractions, even as the muscles adapted and changed over time. These results show that muscle adaptability helped the robot to learn faster and perform more efficiently.

“This brings us one step closer to designing and eventually building biohybrid robots that can adapt to the world around them, just like animals do,” Webster-Wood said.