The up-to-date artificial intelligence-based control system enables cushioned robotic arms to learn a wide repertoire of movements and tasks once, and then adapt to up-to-date scenarios on the fly, without requiring retraining or sacrificing functionality.
This breakthrough brings cushioned robotics closer to human-like adaptability for real-world applications such as assistive robotics, rehabilitation robots, and wearable or medical cushioned robots, making them more knowledgeable, versatile, and unthreatening.
The work was directed by Fr Men’s, Manus and Machina (M3S) interdisciplinary research group – a play on the Latin motto of MIT “mens et manus”, i.e. “mind and hand”, with the addition of “machina” instead of “machine” – as part of Singapore-MIT Alliance for Research and Technology. Co-leads of the project are scientists from the National University of Singapore (NUS) with collaborators from MIT and Nanyang Technological University in Singapore (NTU Singapore).
Unlike regular robots, which move using inflexible motors and joints, cushioned robots are made of malleable materials, such as cushioned rubber, and move using special actuators – elements that act like artificial muscles that cause physical movement. Although their flexibility makes them ideal for exquisite or adaptive tasks, controlling cushioned robots has always been a challenge because their shape changes in unpredictable ways. Real-world environments are often convoluted and full of unexpected disruptions, and even tiny changes in conditions – such as a change in weight, a gust of wind, or a minor equipment failure – can disrupt their performance.
Despite significant advances in cushioned robotics, existing approaches often achieve only one or two of the three functions needed for cushioned robots to operate intelligently in real-world environments: using knowledge acquired from one task to perform another, quickly adapting to changing situations, and ensuring that the robot remains stable and unthreatening by adapting to its movements. The lack of adaptability and reliability has so far been a major obstacle to implementing cushioned robots in real-world applications.
In a publicly available study titled “A generic soft robot controller inspired by the structural and plastic synapses of neurons that adapts to different arms, tasks and perturbations“, published on January 6 in , researchers describe how they developed a new artificial intelligence-based control system that allows soft robots to adapt to a variety of tasks and disturbances. The study draws inspiration from the way the human brain learns and adapts, and was based on extensive research on learning-based robot control, embodied intelligence, soft robotics and meta-learning.
The system uses two complementary sets of “synapses” – connections that regulate how the robot moves – working in tandem. The first set, called “structural synapses,” are trained offline and include a variety of basic movements such as smoothly bending or straightening a soft arm. These represent the robot’s built-in skills and provide a strong, stable foundation. The second set, called “plastic synapses”, constantly updates online as the robot operates, adapting the arm’s behavior to what is happening at that moment. The built-in stability module acts as a safety feature, so even as the robot adjusts during online adaptation, its behavior remains smooth and controlled.
“Soft robots have enormous potential to perform tasks that conventional machines simply cannot perform, but true adoption requires control systems that are both highly efficient and reliably safe. By combining structured learning with real-time adaptability, we have created a system that can handle the complexity of soft materials in unpredictable environments,” says MIT professor Daniela Rus, co-principal investigator of M3S, director of MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), and co-author of the correspondence article. “It’s a step closer to a future where versatile soft robots can operate safely and intelligently alongside humans – in clinics, factories or in everyday life.”
“This up-to-date AI-based control system is one of the first general cushioned robot controllers that can achieve all three key aspects needed for cushioned robots to be used in society and various industries. It can apply the acquired knowledge offline to various tasks, instantly adapt to up-to-date conditions, and remain stable at all times – all within a single control system,” says Associate Professor Zhiqiang Tang, first author and co-author of the corresponding article, who was a postdoctoral fellow at M3S and NUS during the research, and he is currently an associate professor at Southeastern University of China (SEU China).
The system supports multiple task types, allowing the robot’s soft arms to track trajectories, place objects, and adjust the shape of the entire body in one unified approach. The method also allows generalization to various soft-arm platforms, demonstrating its applicability to multiple platforms.
The system was tested and validated on two physical platforms – a cable-driven soft arm and a memory alloy soft actuated arm – and delivered impressive results. Achieved 44-55% reduction in tracking errors in case of strong interference; over 92% shape accuracy against load changes, airflow disruptions and actuator failures; and stable operation even in the event of failure of up to half of the actuators.
“This work redefines what is possible in cushioned robotics. We have shifted the paradigm from tuning and task-specific capabilities towards a truly generalizable framework with human-like intelligence. This is a breakthrough that opens the door to scalable, knowledgeable cushioned machines capable of operating in real-world environments,” says Professor Cecilia Laschi, co-corresponding author and principal investigator at M3S, Vice-Chancellor’s Professor in the NUS Department of Mechanical Engineering in the College of Design and Engineering, and director of the Center Advanced Robotics NUS.
This breakthrough opens the door to more robust, soft robotic systems, enabling the development of manufacturing, logistics, inspection and medical robotics without the need for constant reprogramming, reducing downtime and costs. In healthcare, assistive and rehabilitation devices can automatically adapt their movements to a patient’s changing strength or posture, while wearable or soft medical robots can respond with greater sensitivity to individual needs, improving patient safety and outcomes.
Researchers plan to expand this technology to robotic systems or components that can operate at higher speeds and in more complex environments, with potential applications in assistive robotics, medical devices and industrial soft manipulators, as well as integration with real-world autonomous systems.
Research conducted at SMART was supported by the National Research Foundation of Singapore under the Campus for Research Excellence and Technological Enterprise program.