ROBOTS
Microrobots repair spinal cord
image:
At the start and after three days: the top images show the uninjured spinal cord of a zebrafish; those in the middle show the injured spinal cord; and those at the bottom illustrate how the nerve cells grow thanks to the microrobots.
view moreCredit: ETH Zurich
Spinal cord injuries can have devastating consequences for those affected. Nerve cells in the spinal cord rarely regenerate naturally, while scarring often prevents the regrowth of nerve fibres. Modern therapies attempt to influence implanted stem cells using electrical stimulation to promote the growth of new nerve cells. This approach has several drawbacks: it requires implanted electrodes, and the transplanted cells do not always survive or integrate properly into the existing tissue.
Cells and nanoparticles cleverly combined
Researchers in Zurich are pursuing a new approach, which they have published in the journal Nature Materials. This involves combining therapeutic stem cells with magnetoelectric nanoparticles in such a way that the cells can be guided magnetically to the precise site of an injury and stimulate the stem cells to accelerate repair.
To achieve this, the researchers created a biohybrid microrobot, which combines living neural progenitor cells (NPCs) with a technical component in the form of specially engineered nanoparticles. The NPCs are derived from induced pluripotent stem cells (iPS cells), which are regular body cells reprogrammed in the laboratory to regain stem cell properties. These iPS cells have the potential to differentiate into various types of nervous system cells.
The nanoparticles consist of two layers: an inner layer that responds to magnetic fields and an outer layer that converts this response into electrical signals. By combining these special nanoparticles with the progenitor cells, the researchers fabricate what are known as NPCbots.
A lab the size of a chip
The researchers create the NPCbots in specialised labs on a surface measuring one square centimetre. This process can be illustrated graphically. “We place a reservoir in the centre where we trap the cells. Then we inject the nanoparticles and wait for the two components to bind,” explains Professor Salvador Pané i Vidal of the Multi-Scale Robotics Lab at ETH Zurich.
After just thirty minutes, the NPCbots – each around six micrometres in size – are ready for use. “To scale up fabrication, we operate several lab-on-chip systems in parallel,” explains Hao Ye, senior scientist and the study’s first author. Depending on the test in question, the ETH researchers need hundreds of thousands of microrobots for cell-based studies and several million for animal experiments.
Injured zebrafish swim again
The team tested the NPCbots on zebrafish larvae with spinal cord injuries. The microrobots were injected precisely into the site of the fish’s injury, and electromagnetic fields were generated. For Pané Vidal, teamwork was vital to the experiment’s success: “Stephan Neuhauss and Jingjing Zang at the University of Zurich did extremely valuable work. They enabled us to demonstrate, in a well-characterised regenerative model system, how quickly cells differentiate using our method and how our bots repair the spinal cord.” In just three days, the zebrafish exhibited nearly normal swimming and exploratory behaviour.
The researchers also tested the NPCbots on mice with completely severed spinal cords. Here, too, the results were very promising: after 28 days, the animals’ nerve cells had reconnected at the site of the injury. During this period, the treated mice exhibited increasingly normal movement patterns – their gait, stride length, coordination and exploratory behaviour improved significantly.
This result is particularly significant because, unlike in zebrafish, the mouse spinal cord does not normally regenerate. The treatment was well tolerated by the animals, with no evidence of any adverse effects or immune reactions.
Success through minimally invasive stimulation
These successes were made possible through electrical stimulation of stem cells, greatly enhancing their differentiation after transplantation. In this process, nanoparticles convert magnetic signals directly into electrical impulses that stimulate specific stem cells. When employing NPCbots, researchers need only apply external magnetic fields around the injury site, eliminating the need for implanted electrodes or cables in previous approaches. This is crucial because the spinal cord is extremely sensitive. “Microrobotic guidance makes the treatment more precise and minimally invasive,” Hao explains.
Magnetic fields are particularly well-suited for stimulating stem cells because they can penetrate tissue easily, and their frequency and field strength can be flexibly adjusted to the specific application. Once the progenitor cells have been stimulated and differentiated into nerve cells, the NPCbots essentially dissolve within the tissue. The researchers expect the nanoparticles to be stable and minimally reactive due to their barium titanate coating. Further studies will determine whether and how the particles are degraded or excreted over the long term.
The idea can be expanded as required
The results from animal experiments are extremely promising, but further research will be needed before NPCbots can be tested in humans. “In addition to many clinical aspects, we first need to test which magnetic fields work best in humans and determine the optimal stimulation duration,” Hao explains. Nevertheless, the researchers are already considering further applications: “The reproducible and scalable production of microrobots using our lab-on-a-chip system demonstrates that the platform’s application potential extends beyond basic research,” explains Professor Pané i Vidal. It could also be adapted for other biomedical applications – for example, in cardiology, oncology, wound healing and other targeted regenerative therapies. This could make these treatments safer, more controllable and more effective.
Journal
Nature Materials
Article Title
Magnetoelectric Microrobots for Spinal Cord Injury Regeneration
Article Publication Date
2-Jun-2026
A virtual tomato training arena for harvesting robots
Real-world data create a virtual farm for agricultural AI
Osaka Metropolitan University
image:
The virtual environment (left) is created using real-world tomato farm images (right) manually obtained from camera data collected by agricultural robots.
view moreCredit: Osaka Metropolitan Universit
Researchers at Osaka Metropolitan University have developed a method for creating realistic virtual tomato farms that automatically generate data for training agricultural AI systems. Their approach offers a way to overcome one of the most labor-intensive tasks in farming: harvesting the crops.
Currently, farmbots use object detection systems to locate tomatoes and artificial intelligence to decide whether they are ripe. However, the use of these systems in the field has been bottlenecked by difficulty training them.
AI systems require large amounts of labeled images, which must be gathered from real farms. The process is time consuming as each tomato has to be manually labeled by drawing bounding boxes and assigned a ripeness category. This process is further complicated by natural variations in lighting conditions, plant shapes, and growing environments between each farm and season.
To address this challenge, a research team led by Takuya Fujinaga of the Graduate School of Engineering, Osaka Metropolitan University, created a virtual agricultural environment, which automatically generates realistic tomato images and their corresponding AI training labels.
To make a virtual farm that closely resembled real conditions, the environment was reconstructed using images manually obtained from camera data collected by agricultural robots.
The team used advanced reconstruction methods to build detailed 3D models and Unreal Engine 5 software together with an emerging reconstruction technique known as 3D Gaussian Splatting to reproduce lighting, textures, and geometry. Their virtual agricultural environment created a simulation closer to real-world situations where leaves overlap tomatoes, lighting changes constantly, stems and shadows create clutter, and fruit are often partially hidden.
Using positional information, the system automatically generated labels showing where tomatoes appeared in each image and how ripe they were. The framework also automatically exported annotations in YOLO format, a widely used standard for AI object detection training.
The researchers used the synthetic datasets to train AI models and showed that they could effectively detect tomatoes in real-world images.
“By comparing differences in the shape of 3D tomato models, lighting conditions, and the amount of data, we identified conditions that affect AI accuracy,” Dr. Fujinaga said. “Understanding how lighting, tomato shape, and dataset size affect detection performance are important discoveries for improving the model in the future.”
“Although we studied tomatoes, similar factors are important in harvesting a wide range of agricultural products,” he added. “We are definitely excited about the potential of applying this method to crops other than tomatoes.”
The findings were published in Smart Agricultural Technology.
###
About OMU
Established in Osaka as one of the largest public universities in Japan, Osaka Metropolitan University is committed to shaping the future of society through “Convergence of Knowledge” and the promotion of world-class research. For more research news, visit https://www.omu.ac.jp/en/ and follow us on social media: X, Instagram, LinkedIn.
Journal
Smart Agricultural Technology
Method of Research
Experimental study
Subject of Research
Not applicable
Article Title
Automatic generation of synthetic data and object detection datasets in virtual environments based on tomato-harvesting robot vision
How foundation models will revolutionize robot swarms
image:
Conceptual illustration of a foundation model-enabled robot with an onboard neural network, as envisioned for robot swarm deployment.
view moreCredit: Volker Strobel
Robot swarms are systems composed of many simple robots that coordinate without central control. Soon, they could be radically transformed by artificial intelligence. A new viewpoint article by researchers from the Université Libre de Bruxelles (Belgium) and the CISPA Helmholtz Center for Information Security (Germany) suggests that foundation models—large AI systems trained on vast amounts of data, familiar to many through applications such as ChatGPT—could fundamentally change how robot swarms are designed, deployed, and operated.
Traditionally, robot control software is manually programmed by experts. This process is time-consuming and often inflexible: programmers must anticipate many possible situations in advance, yet real-world deployments can present unexpected events, from robot sensor failures in warehouse operations to the unpredictable conditions that arise during earthquake response.
The viewpoint argues that embedding foundation models into control software could enable robot swarms to achieve levels of autonomy, flexibility, and adaptability that have so far been out of reach. For this purpose, each robot would be equipped with onboard foundation models that process sensor inputs, such as camera images or temperature readings, and generate corresponding collective actions. This could allow swarms to adapt their behavior in real time, deviate from their original tasks when necessary, and interact more naturally with humans through speech or gestures. Consider a robot swarm monitoring a forest that suddenly locates an injured person. Thanks to the foundation model-based control, the swarm could autonomously switch to the more urgent task of providing assistance — not because it was explicitly programmed to do so but because the situation demanded it.
Before this vision can become a reality, swarm robotics research still needs to overcome hardware limitations and better understand how foundation models can translate the behavior of individual robots into coordinated actions at the swarm level. Security also presents a serious concern. For example, hallucinated outputs, where a foundation model generates plausible but incorrect information, could pose significant reliability issues. The researchers therefore advocate a balanced research approach that considers both the possibilities and the associated risks, incorporating them into a comprehensive ethics-by-design framework.
"Foundation models may lay the foundation for robot swarms that autonomously execute responsible actions that consider how humans would react in a similar situation. At the same time, the probabilistic nature of foundation models raises fundamental questions about the trade-off between autonomy and controllability in autonomous systems," says Dr. Volker Strobel, lead author of the article and researcher at IRIDIA (the artificial intelligence lab at the Université Libre de Bruxelles).
Journal
Science Robotics
Subject of Research
Not applicable
Article Title
How Foundation Models Will Revolutionize Robot Swarms
Robot fish could help explain how our ancient ancestors first learned to walk
image:
Researchers have developed a fish-like robot that shows how some species of modern fish are able to walk on land, and could help unravel how early vertebrates evolved similar abilities hundreds of millions of years ago.
view moreCredit: Michael Ishida
Researchers have developed a fish-like robot that shows how some species of modern fish are able to walk on land, and could help unravel how early vertebrates evolved similar abilities hundreds of millions of years ago.
Using a combination of their ‘walking fish’ robot and computer models based on observations of real fish, the researchers, led by the University of Cambridge, found that a wide range of unrelated species have independently evolved the same basic walking gait, which essentially mimics a swimming motion on land.
This simple walking pattern, which the researchers call an ‘undulating tripod gait’, looks flopping and clumsy, but is actually one of life’s most ancient solutions to a problem: how to escape predators or move from one habitat to another, without specialised limbs.
The gait is mechanically simple – fish propel themselves forward with their tails while using their front fins or head for support – and re-emerges in unrelated fish species, from the African lungfish to armoured catfish. Although individual species of walking fish are well-studied, this is the first time that unifying locomotive principles across multiple species have been identified.
This plausible example of convergent evolution – where different species evolve similar abilities independently – could also help researchers understand how vertebrates first made the transition from water to land, one of the most significant events in the history of life on Earth. The results are reported in the journal Nature Communications.
Several species of living fish, including bichirs, lungfish, catfish, sculpin and snakeheads, are capable of walking on land. While they are far more efficient in the water, having an extra mode of locomotion they can use when needed is an evolutionary benefit.
“If you’ve got the ability to walk on land and your predator doesn’t, then you can escape and hopefully the predator moves on,” said lead author Dr Michael Ishida, from Cambridge’s Department of Engineering. “You’ve also got the ability to move from one shallow-water environment to another, like tide pools for example.”
Ishida, an engineer in Professor Fumiya Iida’s lab at Cambridge, worked together with biologists and palaeontologists to study how modern fish walk, and whether those results could be used to help determine how ancient fish made the transition from water to land.
The researchers first created a computer model based on the movements of Polypterus senegalus, a grey bichir native to Africa, and several other walking fish. The model found similar modes of locomotion across several different species.
“We kept seeing this recurring kind of walking motion, although it’s very primitive,” said Ishida. “A number of different fish, spread out across the evolutionary tree, and not closely related to each other, all do it. It’s such a simple movement and can recur from a very basic starting point.”
Ishida and his colleagues called this walking motion an undulating tripod gait: the fish anchors its body with the front fins or head, and uses its tail to push the body forward around that anchor point.
“It looks like a swimming fish dumped onto land,” said Ishida. “A swimming fish uses its body to propel itself through the water, so if you take that, put it on land, give it some ability to shuffle its front fins, that’s exactly what it’s doing.”
The researchers then built a physical robot fish to test their results, and found that the most efficient movement closely matched the bichir’s movements and the results from the computer model.
“We tried all kinds of different gaits on the robot, and every other gait we tried was slower,” said Ishida. “Any time we changed how the body bended, or what sequence it was bended in, it was worse. It was surprising that the optimal walking pattern in the simulation and robot matched what the real fish actually do.”
The researchers say that future work in this area could be applied to fossil fish like Tiktaalik, an important fossil link in the transition from water to land. A similar combination of computer modelling and robotics could help determine how these ancient species first walked on land.
Robot fish could help explain how our ancient ancestors first learned to walk [VIDEO]
Researchers have developed a fish-like robot that shows how some species of modern fish are able to walk on land, and could help unravel how early vertebrates evolved similar abilities hundreds of millions of years ago.
Credit
Michael Ishida
Journal
Nature Communications
Article Title
The undulating tripod gait as a model of the locomotion of walking fish
Article Publication Date
2-Jun-2026
No comments:
Post a Comment