We humans can infer a surprising amount of information—such as emotional fluctuations, level of fatigue, age, or even the presence of physical pain—just by observing someone’s movements, even when those movements are subtle or part of everyday behavior. However, articulating exactly why we sensed those things is often difficult. More often than not, we fall back on vague explanations like “it just felt that way.”
This ability to estimate someone’s physical and emotional state through their movements plays a crucial role in our social lives. Because we can make such inferences unconsciously, we are able to sense others’ feelings, extend help at the right moment, and maintain smooth interpersonal relationships—all without the need for explicit verbal communication.
As robots begin to integrate more deeply into human society and become companions in our daily lives, they too must acquire this ability to “read the room”—that is, to understand others’ internal states from non-verbal cues and respond flexibly. Whether we can create AI that perceives and responds as humans do—this is the central challenge and mission of our research and development efforts.
Humans are able to read each other’s cues because we share a kind of unspoken mutual understanding. This shared understanding is not limited to knowledge gained through language, culture, or education; it may also extend to physical elements such as movement rhythms, structure, and even patterns of muscle use—all of which may be unconsciously shared.
In this research project, we focus on “natural walking” as a manifestation of such unconscious shared understanding. We aim to develop a system that can extract latent issues from a person’s gait and accurately determine when and how robotic intervention may be needed. Our ultimate goal is not simply to process information, but to realize an AI that deeply understands human behavior and condition—and gently points out areas for improvement—so that it can truly live alongside us.

As outlined in the overview, we humans are capable of intuitively sensing another person’s physical or emotional state—such as “they look tired” or “they might have an injury”—just by watching them walk. Aiming to replicate this kind of unconscious observational ability in artificial intelligence, we have carried out our research and development with the key concepts of “human-likeness” and “prediction.”
By the end of the previous fiscal year, we had successfully developed a system that allows AI to observe and analyze natural walking patterns and detect potential health issues or physical irregularities. This system incorporates a mechanism we call the “human-likeness filter,” which enables the AI to predict a person’s immediate future movements based on their current gait. From subtle deviations in this prediction, the system can identify early signs of possible disorders or dysfunctions.
In FY2024, we advanced this technology further by fully automating the system and tackling the challenge of long-term future prediction. Specifically, we developed an AI that can estimate what a person’s physical condition might be 30 years into the future based on their current gait. Moreover, we successfully created an AI capable of predicting how robotic interventions—such as gait support or correction—might alter that future, allowing us to simulate how the projected trajectory of decline or improvement can shift through timely intervention.
Humans, of course, do not consciously control every detail of their movement—like how they swing their arms or step forward—when walking. That is precisely why unconscious elements such as fatigue, emotional state, or pain inevitably show through in one’s gait. These “unconscious bodily expressions” serve as vital clues, revealing internal conditions without words.　As a result of this research, we have now developed an AI capable of answering questions like “What will my walking look like in the future?”, “What kinds of disorders am I at risk of developing?” and “What steps can I take now to prevent that future?”
This technology opens new possibilities for AI that not only understands human movement but also supports the maintenance and improvement of health by looking toward the future together with us.