Soft Robotics x AI
As information processing capabilities of computers improve dramatically, technologies such as large language models, reinforcement learning using real-world simulations, and learning by imitation are advancing at a rapid pace. These technologies are making it possible to generate robot behavior, which was previously extremely difficult, without having to write programs directly. As a result, the burden of manual control and adjustment is being greatly reduced, opening the way for robots to be utilized in a wide variety of environments.
At the same time, however, these AI methods require enormous computational resources, electrical power, and vast amounts of training data, making large-scale computing environments indispensable for learning. In other words, these technologies are “extremely powerful, but also extremely costly.”
In the biological world, the situation is quite different. Despite having limited energy and computational resources (brains), living organisms are able to learn movements suited to their environments and adapt within a short period of time. This remarkable capacity for learning and adaptation is thought to be supported by “morphological computation” - or “embodiment” - arising from their soft and flexible bodies.
Project RAISE, “Real-World Adaptive Intelligence from Soft Embodiment,” seeks inspiration from these biological mechanisms. By actively integrating the properties of soft bodies with AI, the project aims to realize robots that can learn and continuously adapt in real-world environments, much like living organisms. Through the fusion of high-performance AI and the intrinsic “intelligence” provided by soft embodiment, the project envisions a future in which robots can operate more robustly, flexibly, and safely in the real world for extended periods of time.
October 1, 2025
Koh Hosoda (Professor at Kyoto University)
Project Information
Japan Science and Technology Agency (JST)
Strategic Basic Research Program CREST
Research Area “Fundamentals and Core Technologies for Embodied AI”
Project “Real-world Adaptive Intelligence emerging from Soft Embodiment”
（FY 2025-2030, Grant Number: JPMJCR2555）