Dr. Hiroki Mori's research portfolio is expansive, characterized by a deep exploration of how intelligent systems can learn, adapt, and interact with the physical world and humans. His work consistently seeks to imbue robots with more human-like cognitive abilities, focusing on areas crucial for the next generation of AI and robotics. One of Dr. Mori's core research interests lies in "constructive developmental science," which investigates how intelligence emerges and develops through physical embodiment and interaction with the environment. His work extends to analyzing motion, consciousness, and development in infants and even fetuses, using these biological principles to inform the design of intelligent robots. For example, his research on "A human fetus development simulation: Self-organization of behaviors through tactile sensation" explores how basic sensorimotor experiences can lead to complex behavioral patterns, providing insights into the foundational elements of learning in artificial systems. This approach views robots not just as machines to be programmed, but as learning entities that can develop their intelligence through interaction, much like biological organisms. This perspective is particularly relevant for creating adaptive and robust AI systems that can operate in unpredictable real-world environments. In a 2025 publication, he explored "Mathematical Models for Development from fetus to Curiosity," further demonstrating his ongoing commitment to this foundational aspect of AI. Dr. Mori has made significant contributions to the application of deep learning for complex robot manipulation tasks. His research delves into how robots can perceive, understand, and interact with objects and tools in dynamic environments. A prominent example is his work on "In-air knotting of rope using dual-arm robot based on deep learning." This seemingly simple task presents enormous challenges for robots due to the rope's flexibility and constantly changing state. Dr. Mori's team developed a model based on deep neural networks, trained using sensorimotor data, to enable a dual-arm robot to perform bowknots and overhand knots without requiring a dedicated workbench. This research exemplifies the power of deep learning in allowing robots to handle deformable objects, a long-standing hurdle in robotics. Another groundbreaking area is his work on "How to Select and Use Tools? : Active Perception of Target Objects Using Multimodal Deep Learning." This research tackles the critical function of tool selection and use for robots in domestic applications. Dr. Mori's team developed a deep neural network model that learns to recognize object characteristics, acquire tool-object-action relations, and generate motions for tool selection and handling using multimodal data (images, force, and joint angle information). This allows robots to adapt to unknown tools and objects, even generating actions for ingredients transfer tasks with turners or ladles. This represents a significant leap towards robots that can perform versatile tasks requiring nuanced understanding of object properties and functional relationships. His work also extends to "Learning-based collision-free planning on arbitrary optimization criteria in the latent space through cGANs" and "Put-in-Box Task Generated from Multiple Discrete Tasks by a Humanoid Robot Using Deep Learning." These projects aim to equip robots with the ability to navigate complex environments and perform multi-step tasks efficiently and safely, showcasing the practical application of advanced AI techniques in motion planning and task execution. Beyond physical manipulation, Dr. Mori's research also extends into the realm of human-robot interaction, particularly in conversational AI. He has been involved in developing conversational AI for specific, impactful applications, such as supporting patients with dementia. This involves designing socially embodied intelligent agents and robots that can communicate effectively with people, acknowledging the nuances of human conversation, including turn-taking and backchanneling. His work explores how AI agents can facilitate group conversations, a critical function in preventing cognitive decline and providing social support, especially in aging societies. In the rapidly evolving landscape of AI ethics, Dr. Mori has also contributed to discussions on the "Reliability of Artificial Intelligence in Science and Society," addressing critical questions surrounding trust and dependability in AI systems. This demonstrates his holistic approach, considering not just the technical capabilities of AI but also its broader societal implications and ethical considerations. A recurring theme in Dr. Mori's research is predictive learning, where AI models anticipate future states or actions to enable more fluid and adaptive robot behavior. This includes "Deep Predictive Learning for Embodied Intelligence" and its applications in robotics. His work on "Real-Time Motion Generation and Data Augmentation for Grasping Moving Objects with Dynamic Speed and Position Changes" further exemplifies this, allowing robots to robustly interact with objects that are not stationary. By developing models that learn from visual and tactile sensorimotor data, his research enables robots to perform complex tasks with a level of adaptability previously difficult to achieve. This also involves guiding visual attention and using proprioceptive data-driven reinforcement learning for robust task performance under variable conditions.