Soft, robotic cells from morph embed physical AI into hardware

By integrating reinforcement learning with high-fidelity physics-based simulation, morph said it enables a faster translation from concept to product. | Source: morph

As advances in AI have made robots smarter and more capable, some developers are increasingly focusing solely on the software element of intelligence. Robotics startup morph is taking a different approach, one that sees embodied AI as both a hardware and a software problem.

The London-based company embeds sensing and adaptive control directly into reconfigurable deformable materials, enabling real‑time change in morphology and stiffness. The result is soft cells that developers can integrate into a range of robots.

“We’re running real-time physical AI models that can take sensory information and understand it. Then the cells will morphologically change and adapt to affect a change, whether that’s motion, or whether that’s support, or whether that’s protection,” Dr. Jean Nehme, the founder of morph, told The Robot Report . “Whatever it is, we have sensing cells that are soft and able to receive and understand information, and then fundamentally adjust to effect change in shape to benefit a product or a value.”

Inspired by the adaptability of octopuses, morph said it intends to accelerate the development of physical AI systems. Nehme asserted that the key is to bring together AI development and hardware.

morph’s robotic cells have intelligence embedded directly into the materials. This creates cells that can sense , adapt, and respond to the body and the environment in real time, said Nehme.

“I don’t believe that you can build robots with hardware separate, the physical, and AI separate. It just doesn’t work. I think it’s really hard to do right,” he said. “We’re not born with just hardware alone. We’re born with intelligence that learns.”

morph plans to build many different models and deploy them in different ways. In the long term, Nehme said he hopes to use these models to create a more generalizable world model.

“If you watch, whether it’s a puppy or it’s a tiny baby, you watch them learn and adapt. You can see how they’re trying to use a model that helps them think through, ‘How do I grip this?’” Nehme said.

“That integration is super important, and we’re doing that right from the front end,” he added. “We’re able to build design models, deploy models right into different form factors to effectuate change, and then to drop into those models a continuous learning loop. So, when we release something into the wild, it will continue to learn.”