Ryo Ariizumi
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Ryo Ariizumi has a total of 46 co-authors. Here are the top co-authors by number of shared publications:
- Toru Asai
Department of Mechanical Engineering, College of Engineering, Chubu University, Kasugai 487-8501, Japan
16 shared publications - Shun-Ichi Azuma
Kyoto University
15 shared publications - Motoyasu Tanaka
Department of Mechanical and Intelligent Systems Engineering, The University of Electro-Communications, Tokyo, Japan
6 shared publications Shun-Ichi Azuma
Graduate School of Informatics, Kyoto University, Kyoto, Japan
5 shared publicationsToru Asai
Department of Mechanical Engineering, College of Engineering, Chubu University, Kasugai, Japan
5 shared publicationsFumitoshi Matsuno
Department of Electronics and Information Systems Engineering, Osaka Institute of Technology, Osaka, Japan
4 shared publications
Latest Publications
- Open Access
- 12 July 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Dynamic structured pruning and merging for real-time acceleration of reinforcement learning
Deep reinforcement learning (DRL) remains constrained by high computational costs, hindering its practical deployment. While neural network pruning offers a solution for model compression, most existing DRL approaches rely on unstructured pruning. This results in irregular sparse matrices that lack compatibility with standard hardware, offering no tangible real-time acceleration. Furthermore, pruning in DRL is notoriously challenging due to inherent training instability, which often leads to catastrophic performance degradation. To overcome these limitations, we propose a novel framework that integrates dynamic structured pruning with model merging. By periodically merging parallel network instances, our method effectively counteracts the instability triggered by aggressive structural changes, enabling the use of structured pruning directly compatible with general-purpose hardware. Experimental results on continuous control tasks demonstrate that our approach reduces cumulative training FLOPs by up to 72% while maintaining performance competitive with dense baselines across the majority of environments. Additionally, inference profiling confirms an average latency reduction of 16.2%, highlighting the framework’s potential for accelerating both the training and deployment of DRL agents. Our code is available at https://github.com/hail-mary/neuron-pruning
- 12 July 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Open AccessDynamic structured pruning and merging for real-time acceleration of reinforcement learning
Deep reinforcement learning (DRL) remains constrained by high computational costs, hindering its practical deployment. While neural network pruning offers a solution for model compression, most existing DRL approaches rely on unstructured pruning. This results in irregular sparse matrices that lack compatibility with standard hardware, offering no tangible real-time acceleration. Furthermore, pruning in DRL is notoriously challenging due to inherent training instability, which often leads to catastrophic performance degradation. To overcome these limitations, we propose a novel framework that integrates dynamic structured pruning with model merging. By periodically merging parallel network instances, our method effectively counteracts the instability triggered by aggressive structural changes, enabling the use of structured pruning directly compatible with general-purpose hardware. Experimental results on continuous control tasks demonstrate that our approach reduces cumulative training FLOPs by up to 72% while maintaining performance competitive with dense baselines across the majority of environments. Additionally, inference profiling confirms an average latency reduction of 16.2%, highlighting the framework’s potential for accelerating both the training and deployment of DRL agents. Our code is available at https://github.com/hail-mary/neuron-pruning
- 5 January 2026
- Published by Springer Nature in Artificial Life and Robotics
- Vol. 31 (1) , 218-227
Simulation and modular controller integration for self-reconfigurable robots with docking/undocking behaviors
- 5 January 2026
- Published by Springer Nature in Artificial Life and Robotics
- Vol. 31 (1) , 218-227
Simulation and modular controller integration for self-reconfigurable robots with docking/undocking behaviors
- Open Access
- 28 October 2025
- Published by MDPI AG in Mathematics
- Vol. 13 (21) , 3440
Necessary and Sufficient Reservoir Condition for Universal Reservoir Computing
We discuss necessary and sufficient conditions for universal approximation using reservoir computing. Reservoir computing is a machine learning method used to train a dynamical system model by tuning only the static part of the model. The universality is the ability of the model to approximate any dynamical system with any precision. In the previous studies, we provided two sufficient conditions for the universality. We employed the universality definition that has been discussed since the earliest studies on reservoir computing. In this present paper, we prove that these two conditions and the universality are equivalent to one another. Using this equivalence, we show that a universal model must have a “pathological” property that can only be achieved or approached by chaotic reservoirs.
- 28 October 2025
- Published by MDPI AG in Mathematics
- Vol. 13 (21) , 3440
Open AccessNecessary and Sufficient Reservoir Condition for Universal Reservoir Computing
We discuss necessary and sufficient conditions for universal approximation using reservoir computing. Reservoir computing is a machine learning method used to train a dynamical system model by tuning only the static part of the model. The universality is the ability of the model to approximate any dynamical system with any precision. In the previous studies, we provided two sufficient conditions for the universality. We employed the universality definition that has been discussed since the earliest studies on reservoir computing. In this present paper, we prove that these two conditions and the universality are equivalent to one another. Using this equivalence, we show that a universal model must have a “pathological” property that can only be achieved or approached by chaotic reservoirs.
- Open Access
- 13 August 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 29642
Automatic robot design for fault-tolerant robots
It is important for legged robots to be tolerant of joint malfunctions, especially when working in extreme environments such as the surface of an extraterrestrial planet. However, it is not easy to guarantee tolerance, which requires either the robot structure and controller to be carefully designed for specific failure situations or prompt detection and diagnosis of the faults followed by quick controller reconfiguration. To solve this problem, this paper proposes a method using the automatic robot design method recently proposed by some of the authors. The core of the automatic robot design is in the method of updating the robot topology. In this paper, we combine two distinct robot topology update methods to make learning more efficient. Using a physical simulator, we simulate a situation in which several randomly selected joints malfunction. The simulation reveals that the proposed method efficiently trains a robot to walk even when some of its joints are malfunctioning without the need to switch the controller. Test results show that the obtained pair of robot structures and controllers perform well in situations not experienced during training.
- 13 August 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 29642
Open AccessAutomatic robot design for fault-tolerant robots
It is important for legged robots to be tolerant of joint malfunctions, especially when working in extreme environments such as the surface of an extraterrestrial planet. However, it is not easy to guarantee tolerance, which requires either the robot structure and controller to be carefully designed for specific failure situations or prompt detection and diagnosis of the faults followed by quick controller reconfiguration. To solve this problem, this paper proposes a method using the automatic robot design method recently proposed by some of the authors. The core of the automatic robot design is in the method of updating the robot topology. In this paper, we combine two distinct robot topology update methods to make learning more efficient. Using a physical simulator, we simulate a situation in which several randomly selected joints malfunction. The simulation reveals that the proposed method efficiently trains a robot to walk even when some of its joints are malfunctioning without the need to switch the controller. Test results show that the obtained pair of robot structures and controllers perform well in situations not experienced during training.
- Open Access
- 2 May 2025
- Published by Springer Nature in Artificial Life and Robotics
- p. 1-12
Multi-objective path integral policy improvement for learning robotic motion
This paper proposes a new multi-objective reinforcement learning (MORL) algorithm for robotics by extending policy improvement with path integral ( $$\text {PI}^2$$ ) algorithm. For a robot motion acquisition problem, most existing MORL algorithms are hard to apply, because of the high-dimensional and continuous state and action spaces. However, policy-based algorithms such as $$\text {PI}^2$$ can be applied to solve this problem in single-objective cases. Based on the similarity of $$\text {PI}^2$$ and evolution strategies (ESs) and the fact that ESs are well-suited for multi-objective optimization, we propose an extension of $$\text {PI}^2$$ and some techniques to speed up the learning. The effectiveness is shown via numerical simulations.- 2 May 2025
- Published by Springer Nature in Artificial Life and Robotics
- p. 1-12
Open AccessMulti-objective path integral policy improvement for learning robotic motion
This paper proposes a new multi-objective reinforcement learning (MORL) algorithm for robotics by extending policy improvement with path integral ( $$\text {PI}^2$$ ) algorithm. For a robot motion acquisition problem, most existing MORL algorithms are hard to apply, because of the high-dimensional and continuous state and action spaces. However, policy-based algorithms such as $$\text {PI}^2$$ can be applied to solve this problem in single-objective cases. Based on the similarity of $$\text {PI}^2$$ and evolution strategies (ESs) and the fact that ESs are well-suited for multi-objective optimization, we propose an extension of $$\text {PI}^2$$ and some techniques to speed up the learning. The effectiveness is shown via numerical simulations. - 20 December 2024
- Published by Fuji Technology Press Ltd. in Journal of Robotics and Mechatronics
- Vol. 36 (6) , 1357-1368
Motion Design of a Snake Robot to Move Between Two Adjacent Walls
This paper proposes a novel motion design for a snake robot to move freely in a space between two adjacent walls. Snake robots are expected to be useful in several difficult environments because of their high degrees of freedom. Particularly, they are suitable for use in narrow spaces. However, freely moving in narrow spaces, such as spaces between two adjacent walls, remains an open problem. In this study, inspired by helical rolling motion, which is used to move on a pipe, we propose a novel motion method that enables a snake robot to move freely in a narrow space between two adjacent walls. The motion and model are validated using physical simulations.
- 20 December 2024
- Published by Fuji Technology Press Ltd. in Journal of Robotics and Mechatronics
- Vol. 36 (6) , 1357-1368
Motion Design of a Snake Robot to Move Between Two Adjacent Walls
This paper proposes a novel motion design for a snake robot to move freely in a space between two adjacent walls. Snake robots are expected to be useful in several difficult environments because of their high degrees of freedom. Particularly, they are suitable for use in narrow spaces. However, freely moving in narrow spaces, such as spaces between two adjacent walls, remains an open problem. In this study, inspired by helical rolling motion, which is used to move on a pipe, we propose a novel motion method that enables a snake robot to move freely in a narrow space between two adjacent walls. The motion and model are validated using physical simulations.
- Open Access
- 31 October 2024
- Published by Elsevier in International Journal of Electrical Power & Energy Systems
- Vol. 162
Default detection in demand response based on block-sparse structure
- 31 October 2024
- Published by Elsevier in International Journal of Electrical Power & Energy Systems
- Vol. 162
Open AccessDefault detection in demand response based on block-sparse structure
- 9 September 2024
- Published by Springer Nature in International Journal of Control, Automation, and Systems
- Vol. 22 (10) , 3133-3146
The Dynamical Model and the Slithering Controller of Continuous Snakelike Robot on Smooth Surface
- 9 September 2024
- Published by Springer Nature in International Journal of Control, Automation, and Systems
- Vol. 22 (10) , 3133-3146
The Dynamical Model and the Slithering Controller of Continuous Snakelike Robot on Smooth Surface
- 2 August 2024
- Published by Springer Nature in International Journal of Control, Automation, and Systems
- Vol. 22 (8) , 2551-2562
Path-following Control of a Snake Robot With Multiple Passive Joints
- 2 August 2024
- Published by Springer Nature in International Journal of Control, Automation, and Systems
- Vol. 22 (8) , 2551-2562
Path-following Control of a Snake Robot With Multiple Passive Joints
- Open Access
- 22 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Control Systems Letters
- Vol. 8, 826-831
Data Informativity for Distributed Positive Stabilization
This paper investigates data informativity for distributed positive stabilization. Under the situation that the system model is unavailable but the measurement data are available, we address the problem of finding a distributed controller such that the closed-loop system by state feedback is positive and stable. We clarify that a necessary and sufficient condition for the problem to be solvable is characterized by linear matrix inequalities (LMIs), and derive an LMI-based solution to the problem. To demonstrate our results, a numerical example is provided. Moreover, an application to the vehicle formation is given.
- 22 May 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Control Systems Letters
- Vol. 8, 826-831
Open AccessData Informativity for Distributed Positive Stabilization
This paper investigates data informativity for distributed positive stabilization. Under the situation that the system model is unavailable but the measurement data are available, we address the problem of finding a distributed controller such that the closed-loop system by state feedback is positive and stable. We clarify that a necessary and sufficient condition for the problem to be solvable is characterized by linear matrix inequalities (LMIs), and derive an LMI-based solution to the problem. To demonstrate our results, a numerical example is provided. Moreover, an application to the vehicle formation is given.