A robot that changes shape to match its task
In Shenzhen, a team of engineers has drawn from the architecture of living bone to create a machine that refuses the fixed form most robots accept as given. GrowHR, a soft humanoid robot weighing little more than a bag of groceries, can stretch, compress, swim, fly, and walk on water — not as separate tricks, but as expressions of a single unified design. The achievement asks a quiet but consequential question: if a robot can be many things at once, what does that mean for the boundary between tool and organism?
- Most humanoid robots are prisoners of their own rigidity — GrowHR breaks that constraint by extending 315% in height and compressing enough to slip through spaces no conventional machine could enter.
- The tension between softness and strength is resolved through bone-inspired linkages that combine expandable chambers, tensioned cables, and nonstretchable textile — a structural argument borrowed from biology.
- When its linkages and motors synchronize, the robot moves 1,122 times faster than either system alone, a figure that signals how much performance has been left on the table by designing mechanisms in isolation.
- The robot is not merely surviving in difficult environments — it is floating, swimming, flying, and walking on water, each mode a deliberate answer to terrain that defeats single-purpose machines.
- Published in Science Advances, the research is landing as a reframing of what humanoid robotics can aspire to: not a fixed body optimized for one world, but an adaptable form ready for many.
A research team at Southern University of Science and Technology in Shenzhen has unveiled GrowHR, a soft humanoid robot that can fundamentally change its own shape. Weighing just 4.5 kilograms, it stretches to nearly three times its resting height, compresses to navigate narrow gaps, floats, swims, and flies — moving with a fluidity that feels less like machinery and more like living tissue.
The design draws directly from human bone, which is not a simple column but a multifunctional structure combining growth plates, dense and spongy regions, and hollow cavities that balance strength with lightness. GrowHR's linkages replicate this logic: soft expandable chambers, tensioned cables, rigid joint adapters, and a nonstretchable textile wrap that provides axial stiffness. Each linkage weighs just 350 grams.
The performance that follows is striking. The robot can extend its height by 278 percent or shrink 36 percent in height and 61 percent in width. Its coordinated linkage-and-motor system produces crawling speeds 1,122 times faster than either mechanism alone. Fitted with ducted fans or quadrotors, it flies. It also walks across water at 16 millimeters per second — a detail that quietly demonstrates the structural integrity of the whole system.
PhD student Wang Ting noted that stability is maintained through active control algorithms and carbon fiber telescopic linkages, allowing GrowHR to shift between locomotion modes as the environment demands. The researchers envision the robot navigating collapsed buildings, crossing hazardous terrain, assisting in healthcare, and operating in space. Published in Science Advances, the work proposes something larger than a new machine — it proposes a new idea of what a robot is allowed to be.
A team of researchers in Shenzhen has built a robot that does something most machines cannot: it changes shape. GrowHR, unveiled by scientists at Southern University of Science and Technology, is a soft humanoid robot weighing just 4.5 kilograms that can stretch to nearly three times its resting height, compress itself to slip through narrow gaps, float on water, swim, and fly. It moves with a fluidity that feels closer to living tissue than to the rigid metal frames that define most robots today.
The breakthrough lies in how the machine was designed. Rather than copying the simple column-like structures that support most humanoid robots, the team drew inspiration from human bone—specifically, the way bone combines multiple functions in a single structure. Real bone is not uniform. It has growth plates that allow expansion, dense regions for stiffness, spongy sections that absorb impact, and cavities throughout that keep it light without sacrificing strength. GrowHR's linkages mimic this principle. Each one contains soft expandable chambers, tensioned cables running through them, and rigid adapters at the joints. A layer of nonstretchable textile wraps around the whole assembly, providing axial stiffness while a synchronized cable system ensures the extension happens smoothly and evenly. Each linkage weighs just 350 grams.
What this design enables is remarkable. GrowHR can extend its height by 278 percent, reaching 1.36 meters tall, or compress itself by 36 percent in height and 61 percent in width to navigate spaces a rigid robot could never enter. When its linkages and motors work in concert, the robot crawls at a speed 1,122 times faster than either mechanism could achieve alone. It can deliver powerful kicks, storing and releasing energy in ways that rigid robots cannot. The lightweight frame allows it to float and swim. With ducted fans or quadrotors attached, it can fly for distances of several meters. It even walks across water at 16 millimeters per second, a feat that demonstrates both the structural integrity and buoyancy of the design.
Wang Ting, a PhD student at SUST who worked on the project, explained that the robot maintains stability and balance through active control and carbon fiber telescopic linkages. The adaptability comes not just from the hardware but from control algorithms that allow GrowHR to switch between different modes of movement depending on what the environment demands. This is the core innovation: a single robot that can be many things.
The practical applications are substantial. In field rescue operations, GrowHR could navigate through collapsed buildings or wreckage, squeezing through gaps that would trap a conventional robot. Its multiple locomotion modes mean it can move across terrain that would stop other machines—water, narrow passages, open ground. The researchers see potential in hazardous search and rescue, healthcare delivery, space exploration, and any scenario where a robot needs to move through an unstructured, unpredictable environment. The work was published in Science Advances, and it represents a shift in how engineers think about what a humanoid robot can be. Instead of a machine locked into a single form and a fixed set of capabilities, GrowHR suggests a future where robots are as adaptable as the environments they inhabit.
Notable Quotes
The robot manages stability and balance through active control and by adding carbon fiber telescopic linkages.— Wang Ting, PhD student at SUST
This growable bio-inspired structure can be applied in field rescue missions, such as navigating through narrow gaps. Moreover, its multiple locomotion modes allow it to adapt to complex terrains.— Wang Ting, SUST
The Hearth Conversation Another angle on the story
Why does a robot need to look human at all if the real innovation is the bone-inspired linkages?
Because humanoid form opens doors. People trust something that moves like they do. But you're right to push back—the shape matters less than what the shape can do. The human form is just the package for the real breakthrough, which is that this robot can be multiple things.
The numbers are striking—1,122 times faster through coordinated movement. How does that actually work?
Imagine a single motor trying to move the robot forward. Now imagine the linkages extending and contracting in rhythm with that motor, like a person walking. The linkages do the heavy lifting; the motor just coordinates the timing. Together they're exponentially more efficient than either alone.
Walking on water at 16 millimeters per second sounds slow. Why is that impressive?
It's not about speed. It's about what it proves—that the structure is light enough to be buoyant but rigid enough to support its own weight on a surface that would swallow most robots. It's a demonstration of the design's fundamental balance.
What happens when this robot gets deployed in an actual rescue? What could go wrong?
The real test is durability under stress and in unpredictable conditions. A lab is controlled. A collapsed building is not. You don't know what surfaces it will encounter, what forces it will face, whether the cables will fray or the chambers will rupture. That's the next phase.
Is this the beginning of robots that truly adapt, or is it still mostly a clever engineering trick?
It's both. Right now it's a proof of concept—a very elegant one. But the principle it demonstrates, that you can build machines that change their form to match their task, that's foundational. Everything that comes after will build on this.