Researchers at a Chinese university have built a soft, human-inspired robot that can physically grow and shrink on command, mimicking the way human bones elongate during development. The robot, called GrowHR, weighs just 4.5 kg but can increase its height by 36% and its width by 61%, allowing it to crawl, swim, walk on water, and even fly short distances. If the concept scales beyond the lab, it could reshape how engineers think about deploying robots in tight, unpredictable environments where a fixed body simply will not fit.
How Inflatable Bones Let a Robot Grow
Most humanoid robots are locked into a single body shape. GrowHR breaks that constraint with what its creators call “bone-mimetic growable linkages,” a system of inflatable soft chambers held in place by fabric guides that act like the periosteum wrapping real bone. According to the paper hosted on PubMed Central (PMC), these linkages can stretch up to 315% of their resting length. That extensibility is what allows the robot to shift between a compact form and a taller, wider configuration without swapping any hardware. The design draws directly from how growth plates in human long bones add length during childhood, except GrowHR can reverse the process at will, deflating its pneumatic “bones” to shrink and re-inflating them to regain stature.
The team behind the work is led by Associate Professor Wang Hongqiang at the Southern University of Science and Technology in Shenzhen, according to the university’s announcement (SUSTech). Wang’s group demonstrated the robot squeezing through narrow passages by deflating its linkages, then re-inflating to full size on the other side. That shrink-to-pass capability is more than a party trick. In a collapsed building or flooded tunnel, a robot that can flatten itself to navigate debris and then expand to carry tools or sensors would offer a clear advantage over rigid machines that get stuck at the first tight gap. The same growth mechanism also gives engineers a new control dimension: instead of only commanding joint angles and motor speeds, they can tune limb length and torso width to match the task.
Performance Numbers That Stand Out
Raw shape-shifting would mean little if the robot could not actually move well in its different forms. The bibliographic record hosted by the National Library of Medicine at PubMed confirms several striking metrics from the experimental campaign. In its grown configuration, GrowHR crawls 1,122 times faster than a comparable soft-body baseline, a figure that reflects how extending its limbs dramatically improves stride efficiency and ground coverage per cycle. The robot also demonstrated swimming, water-walking, and short-distance flight during testing, highlighting how a single platform can support multiple locomotion modes in the reported experiments. Each mode relies on the same inflatable skeleton, suggesting that a single morphable body can replace multiple specialized platforms.
At 4.5 kg, GrowHR is light enough for a single operator to carry to a deployment site. That low mass is partly a consequence of its pneumatic construction: air-filled chambers weigh far less than the metal actuators and rigid frames found in conventional humanoids. The tradeoff, however, is load capacity. Soft pneumatic systems generally sacrifice force output for compliance, and the published paper does not include head-to-head payload comparisons against rigid competitors. Until those numbers appear, the 1,122-times crawling improvement should be read as a locomotion-speed metric, not a blanket claim of superiority across all tasks. For real-world missions, engineers will have to decide whether GrowHR’s agility and shape-shifting outweigh any limits on how much equipment it can carry or how firmly it can manipulate objects.
Soft Materials and the Safety Question
One argument for building robots out of inflatable, flexible parts is that they are inherently safer around people. A separate preprint on impact-responsive materials explores this idea directly, testing non-Newtonian fluid-based protectors on humanoid frames subjected to repeated falls, drops from 3 m, and stair tumbles. That work, which has not yet been peer-reviewed, suggests soft-bodied robots may absorb impacts differently than rigid counterparts. GrowHR’s inflatable structure could, in principle, help distribute forces over larger areas during collisions, but the GrowHR paper would need dedicated safety testing to substantiate that benefit in collaborative settings.
A related line of research reinforces the value of mixing rigid and soft elements. A peer-reviewed study indexed on PubMed’s database describes a humanoid finger that combines tubular bones, flexible joints, soft skin, and pneumatic actuation, allowing a robotic hand to pinch 0.1 mm paper while also supporting 5.275 kg and grasping fragile items without damage. That finger-level precision hints at what a full-body soft humanoid could eventually achieve if its extremities adopt the same philosophy. GrowHR does not yet demonstrate that level of dexterous manipulation, but its structural approach, layering fabric guides over inflatable cores, follows a compatible design logic. In principle, the same growable linkages that change limb length could be miniaturized to adjust finger span or grip profile on the fly, tailoring contact forces to delicate or irregular objects.
Where the Research Gaps Are
The most candid reading of GrowHR’s current state is that it is a compelling proof of concept with several unanswered engineering questions. The published demonstrations described in the available papers and records indexed via the National Library of Medicine cover controlled lab conditions: flat surfaces, calm water, and pre-set obstacle courses. There is no published data on how the robot performs in mud, rubble, or other chaotic terrain typical of disaster sites, where debris can puncture soft chambers or trap flexible limbs. Energy efficiency data comparing GrowHR to non-growable humanoids is also absent from the available literature, and without those benchmarks it is difficult to judge whether the pneumatic growth system drains batteries faster than a conventional drivetrain would. Because inflation requires compressed air and valves, the onboard power and air-storage architecture may become a limiting factor outside the lab.
Production cost and scalability are similarly unaddressed. Wang Hongqiang’s team has not released figures on what it costs to manufacture the inflatable linkages or how many inflation-deflation cycles they can endure before material fatigue sets in. Those numbers matter because search-and-rescue robots need to survive hundreds of deployments, not just a handful of lab demos. The citation trail visible through personal NCBI collections points to growing academic interest in growable soft structures, but interest alone does not close the gap between a 4.5 kg laboratory prototype and a field-ready system that can be mass-produced, maintained, and repaired under budget constraints. Long-term abrasion, UV exposure, and contamination by dust or chemicals could all degrade the flexible envelopes that make GrowHR work.
From Lab Prototype to Real-World Tool
Bridging that gap will likely involve integrating GrowHR’s morphing skeleton with more mature sensing, control, and manufacturing techniques. For instance, path-planning algorithms could be extended so that the robot not only chooses a route but also optimizes its body size along the way, shrinking to slip through gaps and expanding when stability or reach is more important. Data from related soft-robotics experiments, organized in curated bibliography collections, suggest that combining pneumatic actuators with embedded strain sensors can give soft bodies a form of proprioception. If similar sensing were added to GrowHR’s inflatable linkages, the robot could detect leaks, monitor wear, and adjust inflation to maintain performance as materials age.
Regulatory and ethical considerations will also shape how morphing humanoids leave the lab. Safety standards for collaborative robots currently assume mostly rigid frames with predictable envelopes of motion; a machine that can expand its limbs and torso by more than a third challenges those assumptions. Developers will have to demonstrate that growth behaviors are tightly controlled, fail-safe, and clearly communicated to nearby humans. At the same time, the ability to shrink into a more compact package and then expand to a larger working size could transform logistics for emergency response, remote inspection, and even space missions where volume and mass are at a premium. If future iterations can match the dexterity of rigid hands while retaining GrowHR’s dramatic shape-shifting, the result could be a new class of robots that are not just human-shaped, but human-scaled in a far more literal, adjustable sense.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
