The breakthrough came from a self-training platform that lets lab-grown muscles strengthen themselves around the clock, with no external stimulation.
The same approach could lead to biodegradable environmental sensors, temporary medical implants that dissolve inside the body, and silent underwater drones for defence applications.
In a petri dish at the National University of Singapore, two tiny rings of engineered muscle tissue were pulling against each other like arm-wrestlers locked in a draw. Neither ring would let go. The contractions came from the cells themselves: spontaneous twitches that most researchers had dismissed as a biological curiosity. NUS Assistant Professor Tan Yu Jun saw something else: a free training regimen, running on its own, with no power source, no control unit, and no human intervention.
TIMELINE: Biohybrid Robotics — From Proof of Concept to Record Speed
─────────────────────────────────────────────────────────────────────
2017 ──── 2022 ──── 2024 ──── 2025 ──── 2026
🧪 🧪 🏭 🏭 ◉ NOW
First Biohybrid Biohybrid Neural- OstraBot
biohybrid reviews hand driven record:
actuators catalogue (Waseda) crawler 467mm/min
(Science the field (UIUC) (NUS)
Robotics)
Chronology of key milestones in skeletal muscle-driven biohybrid robotics. Source: Nature Communications, Science Robotics.
The bottleneck that held the field back
Biohybrid robots (machines that combine living muscle tissue with synthetic skeletons) have been a laboratory staple for nearly a decade. The promise is obvious: living actuators are soft, silent, energy-efficient at small scales, and biodegradable. The problem was always force. Most cultured muscle tissues from the standard C2C12 cell line generate less than 1 millinewton of active force. A few labs pushed past 2 mN. The field seemed stuck on a ceiling that nobody knew how to break through.
Researchers tried growth factors. They tried co-culturing muscle cells with photosynthetic algae. They tried electrical stimulation regimes and mechanical stretching. Each method added cost and complexity. None of them changed the fundamental equation: the actuators were too weak to do useful work.
Tan's team took a different route. Instead of designing an elaborate external training apparatus, they built a platform where two muscle tissues train each other.
The arm-wrestling machine
The design is almost childlike in its simplicity. Two rings of engineered skeletal muscle tissue are anchored at one end and connected to a shared sliding block at the other. When one ring contracts, it stretches the other. The stretched ring contracts back. The cycle repeats autonomously throughout the critical maturation window, roughly day three to day five after differentiation, when the cells naturally begin twitching.
The result is continuous mechanical loading that strengthens and aligns both tissues simultaneously. No external power. No control algorithms. No manual intervention. The cells exercise themselves.
After training, the actuators generated 7.05 mN of maximum force with a stress of 8.51 mN/mm², more than three times higher than any previous C2C12-derived muscle actuator reported in the literature. The platform is also scalable: it uses a commercially available cell line (C2C12) and requires no specialized equipment beyond the PDMS molds and sliding block.
Previous best C2C12 muscle actuator: <2 mN
Self-trained actuator (NUS, 2026): 7.05 mN
Improvement: 3.5x over prior art
OstraBot: from stronger muscle to faster robot
The strengthened tissue was integrated into OstraBot, an ostraciiform swimming robot that mimics the stiff-bodied locomotion of boxfish. A single self-trained muscle ring drives two flexible tails through a cable-driven mechanism. A physiology-based contraction model guided the design, identifying stiffness-frequency combinations that maximized the muscle's energy output.
OstraBot reached 467 mm/min, or 15.6 body lengths per minute. The previous record for a skeletal muscle-powered biohybrid robot was roughly 150 mm/min. More importantly, the robot demonstrated precise on-off controllability through sound-triggered clapping control. Past muscle-powered robots either moved constantly without control or were too weak to respond to any signal at all. OstraBot could be told to start and stop.
"Our strengthened skeletal muscle allows the robot to react clearly to an external signal, similar to how nerves control muscles in the body," Tan said. "This demonstrates that biohybrid robots can combine strength with precise regulation, which is essential for real-world applications."
"The purpose of this study was not just to build a faster robot, but to remove a fundamental bottleneck in the field and open the door to high-performance biohybrid systems designed with sustainability in mind."— Asst Prof Tan Yu Jun, National University of Singapore
Where living machines make sense
The question that follows any laboratory breakthrough is where it applies. Biohybrid actuators are never going to replace electric motors in industrial robotics. Their strength-to-weight ratio is still orders of magnitude below electromagnetic drives, and they require a nutrient medium to stay alive. But they have three specific advantages that no conventional motor can match.
First, they are silent. Living muscle contracts without the whine of gears or the hum of windings. For defence applications such as underwater surveillance drones and silent reconnaissance platforms, that is a decisive property. Second, they are biodegradable. An OstraBot variant deployed to monitor a coral reef or a wetland could simply degrade after its mission, leaving no electronic waste. Third, they work at scales where conventional motors become inefficient. Below a certain size threshold, electromagnetic actuators lose power density. Muscle tissue does not.
The field is not waiting for NUS to commercialize the platform. Other groups are moving along parallel tracks. At MIT, Ritu Raman's lab demonstrated artificial hydrogel tendons that boosted biohybrid gripper speed by 3x and force by 30x. At the University of Illinois, the Bashir lab built the first neuromuscular biohybrid crawler, using muscle tissue activated by neurons instead of direct electrical stimulation. At Waseda University, Takeuchi's group developed a biohybrid hand with multi-jointed fingers powered by multiple muscle tissue actuators, or MuMuTAs.
Commercial spinout from NUS or MIT within 12 months
Defence contract for silent underwater biohybrid drone development
Fully autonomous biohybrid robot with onboard nutrient supply
Transition from C2C12 cell lines to primary human muscle cells for higher force output
The shape of the field
Biohybrid robotics is at the stage where clean energy was in the early 2000s. The physics works in the lab, the roadmap is visible, and the question is purely one of engineering and capital. The global biohybrid robots market is projected to reach USD 3 billion by 2030, according to Strategic Market Research. DARPA, the Wellcome Trust, and sovereign wealth funds have all placed early bets.
What changed in 2026 is that the field now has a number. Before OstraBot, running speed was a hope. Now it is 467 mm/min. The next question is whether the NUS self-training platform translates to other muscle types and whether the same arm-wrestling logic can scale from a petri dish to a production line.
For Defence & Robotics readers, the near-term signal is the underwater drone angle. Biohybrid propulsion is silent, biomimetic, and leaves no thermal or electronic signature. A fleet of OstraBot-derived platforms patrolling a subsea cable corridor or a harbour approach would be nearly impossible to detect with conventional acoustic or electromagnetic surveillance. That is not a laboratory hypothetical. It is an engineering problem with a demonstrated actuator platform behind it.
If the same mechanical coupling strategy works for cardiac muscle or primary human cells, the force ceiling of biohybrid actuators could double or triple again within two years. The bottleneck that held the field back for a decade would be gone, and the path from lab demonstration to practical application would shorten dramatically.
The road ahead
Tan's team published the OstraBot results in Nature Communications on March 18, 2026. First author Dr Chen Pengyu won the Best Poster Award for the work at the Materials Research Society Fall Meeting in December 2025. The paper describes a platform that can be replicated in any tissue engineering lab with standard equipment.
What matters is what the platform proves: that living muscle tissue can generate useful mechanical work at a scale where conventional motors struggle, and that the training mechanism is autonomous enough to operate without external intervention. The next step is keeping the tissue alive longer, integrating nutrient delivery, and demonstrating a practical mission: a sensor package carried by an OstraBot variant for 24 hours in a real aquatic environment.
The arm-wrestling machine in the NUS petri dish is still twitching. It has been running continuously since the experiment began.