Yoshiyuki Ohmura

 Yoshiyuki Ohmura.

2007: Completed the Ph.D. program in Department of Mechano Informatics, Graduate School of Information Science and Technology, The University of Tokyo (Information Science and Technology)

2010: Assistant Professor, Graduate School of Information Science and Technology, The University of Tokyo

2015: A  Specially Appointed Researcher at the Graduate School of Information Science and Technology, The University of Tokyo; holds this position to the present.

Inspired by autonomous robots like Astro Boy and Doraemon, I began my research in robotics. Believing that tactile perception is essential for self-awareness, I developed a humanoid robot equipped with full-body tactile sensors during my doctoral studies and conducted research on whole-body movement. Subsequently, to investigate the mechanisms underlying human motion generation, I conducted motion capture studies. Currently, while supervising research on deep imitation learning, I am developing theories to incorporate consciousness, volition, and autonomy into robots.

Affiliation

Hongo 7-3-1, Bunkyo-ku, Tokyo, Japan

The University of Tokyo, Kuniyoshi Lab

A Theory of Consciousness

Science models phenomena using different languages for each theory. What kind of language should the science of consciousness be based on? Physics describes phenomena using terms such as space, time, force, and energy, but the language of physics lacks a way to describe causality. If the science of consciousness is to explore the causes of consciousness, the mechanisms of consciousness, and how artifitial consciousness is designed, I believe the language of physics is insufficient. This is because the concept of the causes of consciousness cannot be addressed within the language of physics and requires the language of causality.

The difference in scientific language lies in the degree of detail with which objects and phenomena are distinguished. A distinction that physics lacks but causality theory possesses is the asymmetry based on manipulability. Let me explain this asymmetry based on manipulability. For example, Newton’s law states that the product of mass and acceleration is equal to force. This law itself contains no asymmetry. However, when considering the actual construction of a mechanism, force is easy to manipulate, whereas acceleration is not. Typically, acceleration is controlled by varying the force. In other words, from the perspective of manipulability, it is natural to view force as the cause and acceleration as the effect. Thus, while the language of physics does not distinguish this asymmetry based on manipulability, this asymmetry becomes essential when considering causal relationships.

When distinguishing between physics and the theory of causation, it becomes clear that many traditional theories of consciousness are constructed using the language of physics, not that of the theory of causation. The theory of consciousness I am developing distinguishes between models based on asymmetries arising from manipulability. In other words, differences between models that cannot be distinguished using the language of physics can be distinguished using the language of the theory of causation. I believe that the cause of consciousness cannot be elucidated without using the language of causality. I feel that this claim is not so much outlandish as it is quite natural, but in traditional theories of consciousness and the philosophy of mind, the distinction between physics and causality has been ambiguous. I believe that Pearl and others have made significant contributions to clarifying this distinction.

In my theory, I define the asymmetry of causality within the brain model itself. The reason for defining it within the brain is that I believe it is the system’s internal state—not its external environment—that determines whether it possesses consciousness. While it is possible to indirectly control whether a system possesses consciousness from the outside—for example, through anesthesia—I believe that, fundamentally, this is controlled within the brain itself, as in the case of sleep and wakefulness. Furthermore, when considering the question of what kind of system possesses consciousness, I think it is natural to assume that the cause lies within the system itself.

Thus, to introduce causal asymmetry within the system, we consider causality across hierarchical levels. Here, “hierarchy” refers to the hierarchical structure of living organisms—a progression from more microscopic structures to more macroscopic ones, such as molecules, cells, cell populations, and organs. In this context, the macroscopic level is formed by a collection of units from the more microscopic level, and the relationship between the parts and the whole is established. We define inter-level causality as causality flowing from the whole level to the part level.

To consider inter-level causality in this way, we must identify asymmetries within the model. In the theory of causation, models are distinguished based on the presence or absence of such asymmetries. In my model (the Dual-Laws Model), I introduce asymmetries between levels based on manipulability. By doing so, I generate new distinctions between models that are not recognized in physics.

Through the lens of physics, all systems change solely through physical forces. However, through the lens of causality theory, we can introduce levels of hierarchy that physics cannot distinguish. In my model, I model mental causation by treating the higher level as the “mental level” and the lower level as the “physical level.” This is made possible by distinguishing between models and levels based on asymmetries in causality, without introducing non-physical forces. In traditional discussions within the philosophy of mind, physics and causality were not clearly distinguished, leading to the conflation of positing non-physical forces with positing mental causation. However, since these are claims within different academic disciplines, they are independent issues. Because this distinction was not made, it was not possible to treat mental causality or the causality of consciousness scientifically. However, if we cannot question the causality of consciousness, I wonder how consciousness can be verified within the physical world.

I believe that mental causation is the cause of consciousness. It is impossible to express this idea using the language of physics. However, I believe that by introducing a new asymmetry based on manipulability, it becomes possible to express this idea without invoking non-physical forces. Furthermore, I believe that it is only by considering theories of consciousness within the framework of causality—rather than physics—that we can begin to think about the cause of consciousness.

Science models phenomena using different languages in each field. I believe that the language of the science of consciousness must model concepts such as the mind, the self, and free will. This is similar to how psychology deals with personality and emotions, and law deals with responsibility and morality. Viewed through the lens of physics, these are concepts that cannot be distinguished. However, there is no necessity to consider the science of consciousness as a branch of physics. I believe that with the appropriate language, it is possible to model the mind, consciousness, and free will. Conversely, if modeling is not possible, I do not fully understand the reason why. However, the ability to model something is distinct from the ability to verify it. The difficulty with consciousness lies in the fact that it is subjective, and conventional scientific verification methods are unlikely to work well for it. However, I do not believe that just because verification is difficult, modeling must also be difficult. My model is, in a sense, surprisingly simple. Please be sure to check out the paper to see the model for yourself.

Imitation Learning

Video Link 

We are also conducting research on imitation learning using deep learning. We have been interested in human motor skills and have developed devices to measure human movement. However, there are significant differences between the human body and a robot’s body, and determining how to map skills has been a major challenge. While our goal was to measure human movements as precisely as possible, we conceived of an approach that was the exact opposite of what had been done before: robotizing the human body. By restricting the human body to the same movements as a robot, it becomes easier to map skills onto the robot. The Tong System is the result of bringing this concept to real-world application.

The basic concept behind the Tong System is that if we measure all the data a human would typically gather while manipulating objects and apply deep learning to it—since we are using exactly the same information—we can narrow the problem down to one of intelligence alone. If there is something humans can do that machines cannot, we can conclude that this represents the limits of intelligence, allowing us to focus solely on the issue of intelligence. It is precisely because we place such importance on "embodiment" that we chose to simplify the issues related to embodiment and concentrate on the problem of intelligence.

The Tongs System uses a tong-shaped hand to sense force in all directions at the fingertips. The information it can capture includes visual, somatosensory, and force data. The operator controls the robot using only the same visual information as the robot via a head-mounted display (HMD). This approach unifies sensory information between the operator and the robot. Another advantage of using an HMD is that it allows for the simultaneous capture of gaze data. When manipulating objects, humans focus their gaze only on the necessary information. In remote-control conditions, in particular, gaze becomes even more critical due to the lack of force feedback.

By utilizing human gaze, we can apply high-resolution imagery only to the area surrounding the line of sight. Thanks to this mechanism, our group’s robots are capable of performing extremely delicate tasks. For example, among the tasks we have successfully accomplished are threading a needle and peeling a banana.

Deep imitation learning is a field that is currently experiencing rapid development, but it fundamentally suffers from the problem that it fails to perform well under conditions that differ significantly from the training data. We are currently researching how robots can understand and perform tasks under such out-of-distribution conditions.

Publications

A Theory of Consciousness

Y.Ohmura, Y.Kuniyoshi (2026) Whole-to-parts causation mechanism, Frontiers in Psychology, 17:1654139. Link

Preprint

Y. Ohmura, Y. Kuniyoshi (2025) Avoiding Epiphenomenalism in Theories of Consciousness: A Causal Framework Based on Asymmetry, arXiv:preprint, arXiv:2511.04047  Link

Y. Ohmura, E.K. Carr, Y. Kuniyoshi (2026) A Mathematical Formalization of Self-Determining Agency, arXiv:preprint, arXiv:2601.02885 Link

Y. Ohmura, Y. Kuniyoshi (2026) Minimal Set of Questions for Theories of Consciousness: Toward a Unified Explanatory Framework, arXiv:preprint, arXiv:2603.12662 Link

Y. Ohmura, Y. Kuniyoshi (2026) Defining causal mechanism in dual process theory and two types of feedback control, arXiv:preprint, arXiv:2602.11478  Link

Y. Ohmura, Y. Kuniyoshi (2026) Causal Stance, arXiv:preprint, arXiv:2604.05004 Link

Y. Ohmura, E.K. Carr, Y. Kuniyoshi (2026) Closure of Self-Determining System Based on Causal and Constitutive Relations, arXiv:preprint, arXiv.2606.21010 Link

Conference

Yoshiyuki Ohmura, Yasuo Kuniyoshi (2026), Any theory of consciousness must explain the causal efficacy of consciousness,  Association for the Scientific Study of Consciousness (ASSC) 29, 2026

Earnest Kota Carr, Yoshiyuki Ohmura,  Moritz Kriegleder, Yasuo Kuniyoshi (2026) ElephantRoom: An Open Platform for Structured Comparison of Consciousness Theories, Association for the Scientific Study of Consciousness (ASSC) 29, 2026 

Yoshiyuki Ohmura, Yasuo Kuniyoshi (2025) What is volition?, Science of Consciousness Conference 

Keiko Fujii, Yoshiyuki Ohmura and Yasuo Kuniyoshi (2018) Synaptic excitatory-inhibitory balance affect information integration via attractor dynamics,

Association for the Scientific Study of Consciousness (ASSC) 22, 2018 

Book

大村吉幸, ロボットの心を創る, Kindle Publishing Link

大村吉幸, 心身問題の科学, Kindle Publishing Link

Imitation Learning

Dataset Link

Ryo Takizawa, Izumi Karino, Koki Nakagawa, Yoshiyuki Ohmura, Yasuo Kuniyoshi (2025) Enhancing Reusability of Learned Skills for Robot Manipulation via Gaze Information and Motion Bottlenecks, IEEE Robotics and Automation Letters, vol. 10, no. 10, pp. 10737-10744 Link 

Heecheol Kim, Yoshiyuki Ohmura, Yasuo Kuniyoshi (2024) Goal-Conditioned Dual-Action Imitation Learning for Dexterous Dual-Arm Robot Manipulation, IEEE Transactions on Robotics, Vol.40, pp. 2287–2305 Link 

Heecheol Kim, Yoshiyuki Ohmura, Akihiko Nagakubo and Yasuo Kuniyoshi (2023) Training Robots Without Robots: Deep Imitation Learning for Master-to-Robot Policy Transfer, IEEE Robotics and Automation Letters, pp. 2906-2913  Link

Heecheol Kim, Yoshiyuki Ohmura, Yasuo Kuniyosh (2021) Gaze-based dual resolution deep imitation learning for high-precision dexterous robot manipulation, IEEE Robotics and Automation Letters, Vol.6, Issue 2, pp.1630-1637  Link  

Heecheol Kim, Yoshiyuki Ohmura, and Yasuo Kuniyoshi (2020) Using Human Gaze to Improve Robustness Against Irrelevant Objects in Robot Manipulation Tasks, IEEE Robotics and Automation Letters, Vol.5, No.3, pp.4415-4422  Link

Ryo Takizawa, Yoshiyuki Ohmura, Yasuo Kuniyoshi (2025) Gaze-Guided Task Decomposition for Imitation Learning in Robotic Manipulation, 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Hangzhou, China, 2025, pp. 7965-7972  Link

H Kim, Y Ohmura, Y Kuniyoshi (2024) Multi-task robot data for dual-arm fine manipulation, 2024 IEEE International Conference on Robotics and Automation (ICRA) Link

Shogo Hamano, Heecheol Kim, Yoshiyuki Ohmura, Yasuo Kuniyoshi (2022) Using human gaze in few-shot imitation learning for robot manipulation,

2022 IEEE/RSJ International Conference on Intelligent Robots and Systems, October 25, 2022 Link

Heecheol Kim, Yoshiyuki Ohmura and Yasuo Kuniyoshi (2022) Memory-based gaze prediction in deep imitation learning for robot manipulation,

2022 IEEE International Conference on Robotics and Automation (ICRA), May 24, 2022  Link

Heecheol Kim, Yoshiyuki Ohmura, Yasuo Kuniyoshi (2021) Transformer-based deep imitation learning for dual-arm robot manipulation,

2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS2021), 2021  Link

Representation Learning

Kayato Nishitsunoi, Yoshiyuki Ohmura, Takayuki Komatsu, Yasuo Kuniyoshi (2025) Learning Conditionally Independent Transformations Using Normal Subgroups in Group Theory, 2025 IEEE International Conference on Development and Learning (ICDL), September 18, 2025.  Link

Takayuki Komatsu, Yoshiyuki Ohmura, Kayato Nishitsunoi, Yasuo Kuniyoshi (2025) Feature-Based Lie Group Transformer for Real-World Applications,

2025 IEEE International Conference on Development and Learning (ICDL), September 18, 2025.  Link

Kayato Nishitsunoi, Yoshiyuki Ohmura, Yasuo Kuniyoshi (2024) Unsupervised Learning for Global and Local Visual Perception Using Navon Figures,

Proceedings of the Annual Meeting of the Cognitive Science Society, Volume 46, pp4841-4847, 2024.  Link

Takayuki Komatsu, Yoshiyuki Ohmura, Yasuo Kuniyoshi (2024) Ablation Study to Clarify the Mechanism of Object Segmentation in Multi-Object Representation Learning,IEEE International Conference on Development and Learning (ICDL), May 21, 2024  Link

Ryo Takatsuki, Yoshiyuki Ohmura, Yasuo Kuniyoshi (2023) Unsupervised Judgment of Properties Based on Transformation Recognition, IEEE International Conference on Development and Learning (ICDL), Macau, China, 2023, pp. 409-414 Link

Takumi Takada, Wataru Shimaya, Yoshiyuki Ohmura, Yasuo Kuniyoshi (2022) Disentangling Patterns and Transformations from One Sequence of Images with Shape-invariant Lie Group Transformer, 2022 IEEE International Conference on Development and Learning (ICDL), September 14, 2022.  Link

Takayuki Komatsu, Yoshiyuki Ohmura, and Yasuo Kuniyoshi (2021) Unsupervised Temporal Segmentation Using Models That Discriminate Between Demonstrations and Unintentional Actions, 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), ThCT5.6, September 30,2021 Link 

Takumi Takada, Yoshiyuki Ohmura, and Yasuo Kuniyoshi (2021) Unsupervised Learning of Shape-invariant Lie Group Transformer by Embedding Ordinary Differential Equation, 2021 IEEE International Conference on Development and Learning (ICDL). IEEE, Beijing China, August 23-26 ,2021.  Link

Tactile Sensor

Takashi Sagisaka, Yoshiyuki Ohmura, Yasuo Kuniyoshi, Akihiko Nagakubo and Kazuyuki Ozaki (2012) Development and applications of High-Density Tactile Sensing Glove, EuroHaptics2012, pp.445–456, 2012  Link

Takashi Sagisaka, Yoshiyuki Ohmura, Akihiko Nagakubo, Yasuo Kuniyoshi (2011) High-density conformable tactile sensing glove, 11th IEEE-RAS International Conference on Humanoid Robots (Humanoids2011), pp.537–542, 2011. Link

Yuki Fujimori, Yoshiyuki Ohmura, Tatsuya harada and Yasuo Kuniyoshi (2009) Wearable Motion Capture Suit with Full-body Tactile Sensors,

2009 IEEE International Conference on Robotics and Automation, pp.3186–3193, 2009.  Link

Yoshiyuki Ohmura, Yasuo Kuniyoshi, Akihiko Nagakubo, Conformable and Scalable Tactile Sensor Skin for Curved Surfaces,

Proc. IEEE Int. Conf. on Robotics and Automation, pp.1348–1353, 2006  Link

Humanoid Robot

Y. Kuniyoshi, Y. Yorozu, S. Suzuki, S. Sangawa, Y. Ohmura, K. Terada and A. Nagakubo (2007) Emergence and Development of Embodied Cognition: A Constructivist Approach Using Robots, Progress in Brain Research, Vol.164, pp.425–445  Link

Yasuo Kuniyoshi, Yoshiyuki Ohmura, Koji Terada, Akihiko Nagakubo, Shinichiro Eitoku, Tomoyuki Yamamoto (2004) Embodied Basis of Invariant Features in Execution and Perception of Whole Body Dynamic Actions — Knacks and Focuses of Roll-and-Rise Motion, Robotics and Autonomous Systems, Vol.48, No.4, pp.189–201 Link

Yasuo Kuniyoshi, Yoshiyuki Ohmura, Koji Terada, Akihiko Nagakubo (2004) Dynamic Roll-And-Rise Motion By An Adult-Size Humanoid Robot, International Journal of Humanoid Robotics, Vol.1, No.3, pp.497–516 Link

Yasuo Kuniyoshi, Yoshiyuki Ohmura, Akihiko Nagakubo (2007) Whole Body Haptics for Augmented Humanoid Task Capabilities, The 13th International Symposium of Robotics Research (ISRR2007), pp.89-100, Hiroshima, Nov. 26, 2007. (Also in: Makoto Kaneko, Yoshihiko Nakamura (eds), Robotics Research –The 13th International Symposium ISRR–, Springer Tracts in Advanced Robotics, vol. 66, ISBN: 978-3-642-14742-5 (Print) 978-3-642-14743-2 (Online), pp. 61-73, 2011.), 2007  Link

Yoshiyuki Ohmura, Yasuo Kuniyoshi (2007) Humanoid Robot Which Can Lift a 30kg Box by Whole Body Contact and Tactile Feedback,

2007 IEEE/RSJ International Conference on Intelligent Robots and Systems, pp.1136–1141, 2007  Link

Yasuo Kuniyoshi, Yoshiyuki Ohmura, Koji Terada, Tomoyuki Yamamoto and Akihiko Nagakubo, Exploiting the Global Dynamics Structure of Whole-Body Humanoid Motion — Getting the Knack of Roll-and-Rise Motion, Proc. Int. Symp. on Robotics Research (ISSR)), 2003. (Also in: Paolo Dario and Raja Chatila (eds), Robotics Research: The Eleventh International Symposium, Springer Tracts in Advanced Robotics, vol. 15, ISBN: 3-540-23214-1, pp. 385-396, 2005.), pp.385–396, 2005.  Link

Koji Terada, Yoshiyuki Ohmura and Yasuo Kuniyoshi (2003) Analysis and Control of Whole Body Dynamic Humanoid Motion — Towards Experiments on a Roll-and-Rise Motion, Proc. IEEE Int. Conf. on Intelligent Robots and Systems, pp.1382–1387, 2003  Link

Yoshiyuki Ohmura, Koji Terada and Yasuo Kuniyoshi (2003) Analysis and Control of Whole Body Dynamic Humanoid Motion — Experiments on a Roll-and-Rise Motion,Proc. Int. Conf. on Humanoid Robotics, 2003. 

Y. Kuniyoshi, Y. Ohmura, K. Terada, A. Nagakubo, S. Eitoku, T. Yamamoto (2004) Embodied Basis of Invariant Features in Execution and Perception of Whole Body Dynamic Actions — Knacks and Focuses of Roll-and-Rise Motion, Robotics and Autonomous Systems, Invited Paper, Vol.48, No.4, pp.189–201, 2004 

Y. Kunioyshi, Y. Ohmura, K. Terada, A. Nagakubo (2004) Dynamic Roll-and-Rise Motion By An Adult-Size Humanoid Robot, International Journal of Humanoid Robotics, Invited Paper, Vol.1, No.3, pp.497–516, 2004. 

Other

Aya Ogasawara, Yoshiyuki Ohmura, Yasuo Kuniyoshi (2020) Reward sensitivity differs depending on global self esteem in value based decision making, Scientific Reports, Vol.10, 21525 Link  

Yoshiyuki Ohmura, Itsuki Ichikawa, Shinichiro Kumagaya, Yasuo Kuniyoshi (2018) Stapedial reflex threshold predicts individual loudness tolerance for people with autistic spectrum disorders, Experimental Brain Research, pp. 1-10  Link   

Yoshiyuki Ohmura, Seiichi Morokuma, Kiyoko Kato, Yasuo Kuniyoshi (2018) Species-specific posture of human foetus in late first trimester, Scientific reports 8(27)  Link 

Yoshiyuki Ohmura, Yasuo Kuniyoshi (2017) A translational model to determine rodent’s age from human foetal age, Scientific Reports 7(17248)  Link 

Yoshiyuki Ohmura, Hirotaka Gima, Hama Watanabe, Gentaro Taga, Yasuo Kuniyoshi (2016) Developmental change in intralimb coordination during spontaneous movements of human infants from 2 to 3 months of age, Experimental Brain Research Vol.234, No.8, pp.2179-2188 Link  

Earnest Kota Carr, Michael Crosscombe, Ahsan Z. Zahid, Yoshiyuki Ohmura, Yasuo Kuniyoshi (2025) Cybernetic Mimesis: Reference Imitation Drives Boom–Bust Cycles, Conference on Artificial Life (ALIFE 2025), October 8, 2025.  Link

Yoshia Abe, Yoshiyuki Ohmura, Shogo Yonekura, Hoshinori Kanazawa and Yasuo Kuniyoshi (2023) Simulating early childhood drawing behaviors under physical constraints using reinforcement learning,IEEE International Conference on Development and Learning (ICDL), Nov 2023  Link

Izumi Karino, Yoshiyuki Ohmura, Yasuo Kuniyoshi (2020) Identifying Critical States by the Action-Based Variance of Expected Return,

International Conference on Artificial Neural Networks (ICANN 2020), 2020.  Link

Kento Sekiya, Yoshiyuki Ohmura and Yasuo Kuniyoshi (2019) Generating an image of an object’s appearance from somatosensory information during haptic exploration, 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS2019), 2019  Link

Katsuma Inoue, Yoshiyuki Ohmura, Shogo Yonekura, Yasuo Kuniyoshi (2017) The contribution of topology for inclusion of feedforward network and biased synaptic strength to the long-term memory effect in a cortical microcircuit, in Proceedings of The 26th Annual Computational Neuroscience Meeting (CNS2017), 2017. 

Kunihiro Ogata, Daisuke Shiramatsu, Yoshiyuki Ohmura and Yasuo Kuniyoshi (2009) Analyzing the “Knack” of Human Piggyback Motion Based on Simultaneous Measurement of Tactile and Movement Data as a Basis for Humanoid Control, The 2009 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS2009), pp.2531–2536, 2009  Link

Earnest Kota Carr, Yoshiyuki Ohmura, Yasuo Kuniyoshi (2025) A Computational Model of Perceptual Assimilation: Algebraic Structure Feedback Control and Genetic Epistemology, Jean Piaget Society