Researchers have created an AI-based control system that allows soft robotic arms to learn a broad range of movements and tasks a single time, then adapt in real time to new conditions without retraining or loss of performance.
This advance pushes soft robotics toward human-like flexibility in real-world uses—such as assistive and rehabilitation robotics, wearable devices, and medical soft robots—by improving intelligence, versatility, and safety.
The work was led by the Singapore-MIT Alliance for Research and Technology’s (SMART) Mens, Manus & Machina (M3S) interdisciplinary group and the National University of Singapore (NUS), with collaborators from the Massachusetts Institute of Technology (MIT) and Nanyang Technological University (NTU Singapore).
Unlike conventional robots, soft robots use flexible materials and muscle-like actuators instead of rigid joints and motors. Their flexibility suits delicate tasks but makes control difficult due to unpredictable shape changes. In real-world settings, minor changes like added weight, wind, or hardware issues can disrupt their motion.
The Need for Advanced Control in Soft Robots
Current soft robotics methods usually achieve only one or two of three key abilities: task transfer, adaptation, and stability. This gap in adaptability and reliability has limited the deployment of soft robots outside the lab.
A recent Science Advances study introduces an AI system that lets soft robots adapt to diverse tasks and disturbances. The system, inspired by brain learning, adapts using advances in robotic control, soft robotics, and meta-learning.
Mechanism of the Brain-Inspired Controller
The system relies on two complementary types of “synapses”—connections that modulate the robot’s movements—operating together. The first type, “structural synapses,” is trained offline on basic motions like bending or extending a soft arm. These create the robot’s core skills, giving it a stable and reliable foundation.
The second type, “plastic synapses,” updates in real time to adjust the arm’s movements to current conditions. A built-in stability mechanism keeps the robot’s movements smooth and controlled during online adaptation.
AI Controller Powers Safe, Adaptive Soft Robots
“This AI control system is among the first general soft-robot controllers to achieve all three essential capabilities for practical use: transferring learned skills across tasks, adapting instantly to new conditions, and maintaining stability—all within a single framework,” said Associate Professor Zhiqiang Tang, who conducted the research as a Postdoctoral Associate at M3S and NUS, is the first and co-corresponding author of the study, and is now Associate Professor at Southeast University (SEU China).
“Soft robots have tremendous potential to perform tasks beyond the reach of conventional machines, but their widespread adoption depends on control systems that are both highly capable and reliably safe. By combining structural learning with real-time adaptability, we’ve developed a system that can manage the complexity of soft materials in unpredictable environments. This brings us closer to a future where versatile soft robots can operate safely and intelligently alongside humans—in clinics, factories, and daily life,” said Professor Daniela Rus, Co-lead Principal Investigator at M3S, Director of MIT’s CSAIL, and co-corresponding author of the paper.
Practical Trials and Future Directions
The system handles multiple tasks, enabling soft arms to track trajectories, place objects, and regulate shape in one approach. It also generalizes across different soft-arm platforms, showing strong cross-platform versatility.
The researchers tested the system on two soft-arm platforms with impressive results. It reduced tracking errors by 44–55%, maintained over 92% shape accuracy, and remained stable despite half the actuators failing.
“This work redefines the possibilities in soft robotics. We’ve moved from task-specific tuning to a truly generalizable framework with human-like intelligence. “This breakthrough allows intelligent soft robots to operate reliably in real-world settings,” said Professor Cecilia Laschi, M3S Principal Investigator.
This breakthrough enables resilient soft robots for industry and healthcare, cutting reprogramming, downtime, and costs. In healthcare, soft robots can adapt to patients’ needs, improving safety and outcomes.
Researchers plan to extend this technology to faster, more complex robots for healthcare, industry, and autonomous systems.
Read the original article on: Tech Xplore
Read more: She Moves Naturally, Expresses Emotion, Maintains Eye Contact, and Feels Warm—Yet She’s a Robot