Kazuhiro Ohkura
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
Kazuhiro Ohkura has a total of 55 co-authors. Here are the top co-authors by number of shared publications:
- Motoaki Hiraga
Kyoto Institute of Technology, Goshokaido-Cho, Matsugasaki, Sakyo-ku, Kyoto 606-8585, Japan
14 shared publications - Toshiyuki Yasuda
Faculty of Engineering, University of Toyama, Toyama, Japan
8 shared publications - Daichi Morimoto
Hiroshima University, Hiroshima, Japan
5 shared publications Masaharu Munetomo
Information Initiative Center, Hokkaido University, Hokkaido, Japan
4 shared publications- Xiaotong Nie
Graduate School of Advanced Science and Engineering, Hiroshima University, Hiroshima, Japan
4 shared publications - Yoshiaki Katada
Department of Electrical and Electronic Engineering, Setsunan University, Neyagawa, Japan
4 shared publications
Latest Publications
- Open Access
- 15 July 2024
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 19 (7) , e0307084
Improving the performance of mutation-based evolving artificial neural networks with self-adaptive mutations
Neuroevolution is a promising approach for designing artificial neural networks using an evolutionary algorithm. Unlike recent trending methods that rely on gradient-based algorithms, neuroevolution can simultaneously evolve the topology and weights of neural networks. In neuroevolution with topological evolution, handling crossover is challenging because of the competing conventions problem. Mutation-based evolving artificial neural network is an alternative topology and weights neuroevolution approach that omits crossover and uses only mutations for genetic variation. This study enhances the performance of mutation-based evolving artificial neural network in two ways. First, the mutation step size controlling the magnitude of the parameter perturbation is automatically adjusted by a self-adaptive mutation mechanism, enabling a balance between exploration and exploitation during the evolution process. Second, the structural mutation probabilities are automatically adjusted depending on the network size, preventing excessive expansion of the topology. The proposed methods are compared with conventional neuroevolution algorithms using locomotion tasks provided in the OpenAI Gym benchmarks. The results demonstrate that the proposed methods with the self-adaptive mutation mechanism can achieve better performance. In addition, the adjustment of structural mutation probabilities can mitigate topological bloat while maintaining performance.
- 15 July 2024
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 19 (7) , e0307084
Open AccessImproving the performance of mutation-based evolving artificial neural networks with self-adaptive mutations
Neuroevolution is a promising approach for designing artificial neural networks using an evolutionary algorithm. Unlike recent trending methods that rely on gradient-based algorithms, neuroevolution can simultaneously evolve the topology and weights of neural networks. In neuroevolution with topological evolution, handling crossover is challenging because of the competing conventions problem. Mutation-based evolving artificial neural network is an alternative topology and weights neuroevolution approach that omits crossover and uses only mutations for genetic variation. This study enhances the performance of mutation-based evolving artificial neural network in two ways. First, the mutation step size controlling the magnitude of the parameter perturbation is automatically adjusted by a self-adaptive mutation mechanism, enabling a balance between exploration and exploitation during the evolution process. Second, the structural mutation probabilities are automatically adjusted depending on the network size, preventing excessive expansion of the topology. The proposed methods are compared with conventional neuroevolution algorithms using locomotion tasks provided in the OpenAI Gym benchmarks. The results demonstrate that the proposed methods with the self-adaptive mutation mechanism can achieve better performance. In addition, the adjustment of structural mutation probabilities can mitigate topological bloat while maintaining performance.
- Open Access
- 20 June 2024
- Published by Fuji Technology Press Ltd. in Journal of Robotics and Mechatronics
- Vol. 36 (3) , 538-545
Collective Transport Behavior in a Robotic Swarm with Hierarchical Imitation Learning
Title: Collective Transport Behavior in a Robotic Swarm with Hierarchical Imitation Learning | Keywords: swarm robotics, automatic design, reinforcement learning, imitation learning | Author: Ziyao Han, Fan Yi, and Kazuhiro Ohkura
- 20 June 2024
- Published by Fuji Technology Press Ltd. in Journal of Robotics and Mechatronics
- Vol. 36 (3) , 538-545
Open AccessCollective Transport Behavior in a Robotic Swarm with Hierarchical Imitation Learning
Title: Collective Transport Behavior in a Robotic Swarm with Hierarchical Imitation Learning | Keywords: swarm robotics, automatic design, reinforcement learning, imitation learning | Author: Ziyao Han, Fan Yi, and Kazuhiro Ohkura
- 21 October 2023
- Published by Springer Nature in Artificial Life and Robotics
- Vol. 28 (4) , 661-668
An evolutionary robotics approach to a multi-legged robotic swarm in a rough terrain environment
- 21 October 2023
- Published by Springer Nature in Artificial Life and Robotics
- Vol. 28 (4) , 661-668
An evolutionary robotics approach to a multi-legged robotic swarm in a rough terrain environment
- 17 October 2023
- Published by Springer Nature in Artificial Life and Robotics
- Vol. 28 (4) , 703-709
Developing multi-agent adversarial environment using reinforcement learning and imitation learning
- 17 October 2023
- Published by Springer Nature in Artificial Life and Robotics
- Vol. 28 (4) , 703-709
Developing multi-agent adversarial environment using reinforcement learning and imitation learning
- 5 October 2023
- Published by Springer Nature in Artificial Life and Robotics
- Vol. 28 (4) , 669-679
Generating collective behavior of a robotic swarm using an attention agent with deep neuroevolution
- 5 October 2023
- Published by Springer Nature in Artificial Life and Robotics
- Vol. 28 (4) , 669-679
Generating collective behavior of a robotic swarm using an attention agent with deep neuroevolution
- Open Access
- 20 August 2023
- Published by Fuji Technology Press Ltd. in Journal of Robotics and Mechatronics
- Vol. 35 (4) , 977-987
Generating Collective Behavior of a Multi-Legged Robotic Swarm Using Deep Reinforcement Learning
This paper presents a method of generating collective behavior of a multi-legged robotic swarm using deep reinforcement learning. Most studies in swarm robotics have used mobile robots driven by wheels. These robots can operate only on relatively flat surfaces. In this study, a multi-legged robotic swarm was employed to generate collective behavior not only on a flat field but also on rough terrain fields. However, designing a controller for a multi-legged robotic swarm becomes a challenging problem because it has a large number of actuators than wheeled-mobile robots. This paper applied deep reinforcement learning to designing a controller. The proximal policy optimization (PPO) algorithm was utilized to train the robot controller. The controller was trained through the task that required robots to walk and form a line. The results of computer simulations showed that the PPO led to the successful design of controllers for a multi-legged robotic swarm in flat and rough terrains.
- 20 August 2023
- Published by Fuji Technology Press Ltd. in Journal of Robotics and Mechatronics
- Vol. 35 (4) , 977-987
Open AccessGenerating Collective Behavior of a Multi-Legged Robotic Swarm Using Deep Reinforcement Learning
This paper presents a method of generating collective behavior of a multi-legged robotic swarm using deep reinforcement learning. Most studies in swarm robotics have used mobile robots driven by wheels. These robots can operate only on relatively flat surfaces. In this study, a multi-legged robotic swarm was employed to generate collective behavior not only on a flat field but also on rough terrain fields. However, designing a controller for a multi-legged robotic swarm becomes a challenging problem because it has a large number of actuators than wheeled-mobile robots. This paper applied deep reinforcement learning to designing a controller. The proximal policy optimization (PPO) algorithm was utilized to train the robot controller. The controller was trained through the task that required robots to walk and form a line. The results of computer simulations showed that the PPO led to the successful design of controllers for a multi-legged robotic swarm in flat and rough terrains.
- Open Access
- 20 August 2023
- Published by Fuji Technology Press Ltd. in Journal of Robotics and Mechatronics
- Vol. 35 (4) , 988-996
When Less Is More in Embodied Evolution: Robotic Swarms Have Better Evolvability with Constrained Communication
Title: When Less Is More in Embodied Evolution: Robotic Swarms Have Better Evolvability with Constrained Communication | Keywords: embodied evolution, swarm robotics, evolutionary robotics, collective adaptive systems | Author: Motoaki Hiraga, Daichi Morimoto, Yoshiaki Katada, and Kazuhiro Ohkura
- 20 August 2023
- Published by Fuji Technology Press Ltd. in Journal of Robotics and Mechatronics
- Vol. 35 (4) , 988-996
Open AccessWhen Less Is More in Embodied Evolution: Robotic Swarms Have Better Evolvability with Constrained Communication
Title: When Less Is More in Embodied Evolution: Robotic Swarms Have Better Evolvability with Constrained Communication | Keywords: embodied evolution, swarm robotics, evolutionary robotics, collective adaptive systems | Author: Motoaki Hiraga, Daichi Morimoto, Yoshiaki Katada, and Kazuhiro Ohkura
- Open Access
- 20 August 2023
- Published by Fuji Technology Press Ltd. in Journal of Robotics and Mechatronics
- Vol. 35 (4) , 997-1006
MBEANN for Robotic Swarm Controller Design and the Behavior Analysis for Cooperative Transport
This study focuses on mutation-based evolving artificial neural network (MBEANN), a topology and weight evolving artificial neural network (TWEANN) algorithm. TWEANN optimizes both the connection weights and neural network structure. Primarily, MBEANN uses only mutations to evolve artificial neural networks. An individual in an MBEANN is designed to have a set of sub-networks called operons. Operons are expected to have functions during evolution because they do not recombine with other operons. In this study, we applied MBEANN to design a controller for a robotic swarm on cooperative transport, where the following canonical evolving artificial neural network (EANN) methods do not work well. For comparison with MBEANN, we used an EANN with a fixed network structure and neuroevolution of augmenting topologies (NEAT), which is a widely used TWEANN algorithm. We confirmed that the robot controller that evolved with the MBEANN outperformed the structure-fixed EANN and NEAT controllers. In addition, we investigated the behavior of the swarm robot obtained using the proposed method, in which we deactivated each operon to extract its function. The results show that operons could have their functions, and that several operons could strengthen one another’s functions.
- 20 August 2023
- Published by Fuji Technology Press Ltd. in Journal of Robotics and Mechatronics
- Vol. 35 (4) , 997-1006
Open AccessMBEANN for Robotic Swarm Controller Design and the Behavior Analysis for Cooperative Transport
This study focuses on mutation-based evolving artificial neural network (MBEANN), a topology and weight evolving artificial neural network (TWEANN) algorithm. TWEANN optimizes both the connection weights and neural network structure. Primarily, MBEANN uses only mutations to evolve artificial neural networks. An individual in an MBEANN is designed to have a set of sub-networks called operons. Operons are expected to have functions during evolution because they do not recombine with other operons. In this study, we applied MBEANN to design a controller for a robotic swarm on cooperative transport, where the following canonical evolving artificial neural network (EANN) methods do not work well. For comparison with MBEANN, we used an EANN with a fixed network structure and neuroevolution of augmenting topologies (NEAT), which is a widely used TWEANN algorithm. We confirmed that the robot controller that evolved with the MBEANN outperformed the structure-fixed EANN and NEAT controllers. In addition, we investigated the behavior of the swarm robot obtained using the proposed method, in which we deactivated each operon to extract its function. The results show that operons could have their functions, and that several operons could strengthen one another’s functions.
- 26 December 2022
- Published by Springer Nature in Artificial Life and Robotics
- Vol. 28 (1) , 158-165
Autonomous highway driving using reinforcement learning with safety check system based on time-to-collision
- 26 December 2022
- Published by Springer Nature in Artificial Life and Robotics
- Vol. 28 (1) , 158-165
Autonomous highway driving using reinforcement learning with safety check system based on time-to-collision
- 29 November 2022
- Published by Springer Nature in Artificial Life and Robotics
- Vol. 28 (1) , 67-75
Generating collective wall-jumping behavior for a robotic swarm with self-teaching automatic curriculum learning
- 29 November 2022
- Published by Springer Nature in Artificial Life and Robotics
- Vol. 28 (1) , 67-75
Generating collective wall-jumping behavior for a robotic swarm with self-teaching automatic curriculum learning