Breakthrough AI Control System Empowers Adaptive Soft Robots

A novel artificial intelligence (AI) control system has been developed to enable soft robotic arms to learn various motions and tasks, then adapt to new situations without requiring retraining. This development aims to improve the adaptability of soft robots for real-world applications such as assistive robotics, rehabilitation robots, and medical devices.

The research was conducted by the Mens, Manus and Machina (M3S) interdisciplinary research group within the Singapore-MIT Alliance for Research and Technology. Collaborators included the National University of Singapore (NUS), MIT, and Nanyang Technological University in Singapore (NTU Singapore).

The Challenge of Soft Robotics

Unlike rigid robots, soft robots are constructed from flexible materials and utilize actuators for movement. The inherent flexibility of these robots, while beneficial for delicate tasks, presents significant challenges in control due to their unpredictable shape changes. Real-world disturbances, such as shifts in weight or hardware faults, can disrupt their movements. Previous approaches often lacked the combined capabilities of transferring learned skills, adapting quickly to changes, and ensuring stability.

Published in Science Advances, the study details an AI control system that allows soft robots to adapt across diverse tasks and disturbances. The system draws inspiration from human brain learning and adaptation processes.

System Mechanism

The control system employs two sets of "synapses" that adjust robot movement:

• Structural Synapses: These are trained offline on foundational movements, providing a stable base of built-in skills.
• Plastic Synapses: These update continuously online as the robot operates, fine-tuning its behavior in real-time.

A built-in stability measure ensures smooth and controlled adaptation.

"The system addresses the complexity of soft materials in unpredictable environments by combining structural learning with real-time adaptiveness."
— Professor Daniela Rus, co-lead principal investigator at M3S and director of the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL)

"This AI control system is among the first to integrate offline learning transfer, instant adaptation, and stability within a single control framework for soft robots."
— Associate Professor Zhiqiang Tang, first author of the paper

Capabilities and Validation

The system supports multiple task types, allowing soft robotic arms to perform trajectory tracking, object placement, and whole-body shape regulation. It also applies across different soft-arm platforms.

Tests on cable-driven and shape-memory-alloy-actuated soft arms demonstrated:

• 44-55 percent reduction in tracking error under significant disturbances.
• Over 92 percent shape accuracy despite payload changes, airflow disturbances, and actuator failures.
• Stable performance even with up to half of the actuators failing.

"This work shifts the paradigm towards a generalizable framework for intelligent soft machines."
— Professor Cecilia Laschi, co-corresponding author and principal investigator at M3S

Future Applications

This technology aims to facilitate the development of more robust soft robotic systems for manufacturing, logistics, inspection, and medical robotics, potentially reducing reprogramming needs, downtime, and costs. In healthcare, assistive and rehabilitation devices could automatically adjust to patient needs, while wearable and medical soft robots could respond more sensitively, enhancing safety and patient outcomes.

Researchers plan to expand this technology to systems operating at higher speeds and in more complex environments. The research received support from the National Research Foundation Singapore.