OSX Research Group Introduction Vol. 1: ROBOTICS Group – Soft Robotics –

　We present a series of articles introducing the OSX Research Group. In the first installment, we provide an overview of the Soft Robotics technology developed by the ROBOTICS Group.

　Our goal is to achieve robots capable of performing various tasks, such as component assembly and cooking, in the same space as humans, fostering harmony between humans and machines. Traditional robots have been required to perform predetermined actions quickly and accurately. To execute these predetermined actions, meticulous preparation of the work environment specific to the robot was necessary. However, when working in the same space as humans, it is not always possible to prepare a work environment tailored to the robot, and the requirements of the tasks can vary depending on the situation. Consequently, relying solely on predetermined actions may not be sufficient. On the other hand, it is also extremely difficult to design robotic actions manually to account for every possible situation. Therefore, there is a need for robots to adapt immediately to newly encountered, unprepared environments while ensuring safety with few instructions.¹

^{1: Example of a prepared environment: For instance, in the case of component assembly, it is necessary to design fixtures to grip the target components accurately, perform position alignment, and accurately sense the assembly position of the components.}

　We are developing soft robots that combine physically flexible bodies with machine learning. To adapt effectively to unprepared new environments, robots need to autonomously explore the environment and acquire methods for task completion through trial and error. However, trial and error in the real world poses challenges involving risks. Physically flexible bodies enable robots to explore the environment safely by conforming to the work environment through deformation of their flexible parts, allowing for trial and error. To enable a wide range of tasks, we are developing lightweight and flexible tendon-driven robots, as well as flexible wrist modules that can be easily integrated into commonly used rigid robots. Additionally, we are developing machine learning and reinforcement learning techniques that allow tasks to be accomplished with a minimal number of trials and enable rapid adaptation to new environments by leveraging past experiences.

　We will accelerate soft robots’ integration into society by leveraging the synergies between soft bodies and machine learning.