In order to realize an endoscopic surgery support AI Robot with decision-making assistance features, we are advancing research from the perspective of kinematic control. In the endoscopic surgery support AI Robot, taking into account the ambiguity inherent in machine learning, various supportive information will be provided. Movements, such as the operation of the endoscopic camera and auxiliary forceps, demand continuous awareness of the motions of cameras and forceps designed for endoscopic surgery. Based on this, it is required to preemptively offer the information that the doctor wants, thereby assisting the physician. The assistance information considered here is not limited to visual data but also contemplates representation in the robot's movements. By incorporating ambiguity in presenting this information, users are prompted to make decisions proactively. We believe that through this approach, AI can earn trust from humans. 

We are undertaking the development of the foundational technology for Reliable Interventional AI Robotics, which forms the core and takes into account ambiguity. Here, we are working on the development of generative learning methods designed to output information essential for decision support from a comparatively small set of labeled data. Additionally, we are researching mathematical and computational frameworks to realize machine generative learning that leverages robot motion actions. 

We are working on the development of a micro-manipulation support AI Robot that uses Reliable Interventional AI as its foundation, ensuring trustworthy assistance in intricate manipulations. Based on the inference results obtained from Reliable Interventional AI, we assist in micro-manipulation tasks. For instance, in intracytoplasmic sperm injection (ICSI) as part of infertility treatments, embryologists handle embryos, and it's essential to manipulate them with the fewest touches possible to prevent damage. In such delicate tasks, our system constantly monitors microscope images, movements of the micromanipulator, and the operator's actions. Based on this data, the Reliable Interventional AI suggests appropriate manipulation points, incorporating inherent ambiguity. The proposed appropriate operation trajectory is expressed probabilistically, and a system determining the force distribution between the operator and the assisting robot based on that probability is under development. Moreover, while collecting sufficient labeled data for micro-manipulation is challenging, creating an environment to acquire data from the human-machine system that allows mechanical repetition is also a focal area of our research .