Genetic Programming (GP) is an evolutionary computation method that optimizes the rules defining the relationship between environmental states and system output. GP is an effective method for dynamic environments in which the information repeatedly changes multiple times. On the other hand, in GP, the rules are evaluved as the environmental state changes, so the rules acquired in the distant past disappear in time, and re-learning is required in a dynamic environment. This paper proposes an optimization method for the dynamic scheduling problem where new jobs arrive intermittently. Specifically, a method to improve the learning efficiency of GP in such a periodic dynamic environment by dividing the population into several subpopulations and recording the environmental states or their characteristics. This paper conducts some numerical experiments on the dynamic scheduling problems in which new jobs arrive irregularly to verify the usefulness of the proposed method.
Domain adaptation is an approach to transfer knowledge from a certain source domain to another target domain. Several studies are recently reported on partial domain adaptation (PDA), where the class set of the source domain is larger than that of the target domain as a more realistic setting, but then the source domain specific classes make the adaptation more difficult. Most existing methods for PDA give small weights to the source domain specific classes to prevent the target data from being matched. The present paper proposes a PDA method which introduces a novel mechanism that gives additional weights to an individual target data by estimating the probability that the data belongs to each source class. The estimation is given by multiple discriminators that measure the distance between the data distribution of each source class and the entire target data distribution through adversarial training against a data encoder. Computer experiments using two handwritten digit datasets as two domains show that the proposed method achieves more stable and accurate domain adaptation compared with state-of-the-art existing methods for PDA.
In order to realize autonomous drones that collect information and transport supplies by air, it is required to automatically detect a safe landing site in an unknown environment. In this paper, we propose a method to find a candidate landing site using ground images captured by a monocular camera from a drone in flight. The proposed method evaluates the safety of the ground surface by combining the surface classification through Semantic Segmentation and the flatness estimation from dense optical flow. The evaluation is performed for each pixel of the captured images, and a detailed shape of the possible landing area can be obtained. We applied the method to actual images taken by a drone and verified that the landable area was extracted from an altitude of about 100 meters.
Optimization in layouts and robot motions is an essential yet challenging task for robotic cell systems in crowded spaces. In the conventional research, since the robot posture is not explicitly considered, there may be no solution or the operation time may be long if the motion is optimized for the layout obtained by the arrangement optimization. In this work, focusing on the importance of robot posture, we propose a hierarchical optimization framework that newly adds posture optimization between placement optimization and motion optimization. Furthermore, when solving the placement optimization with a genetic algorithm, we devised an objective function in order to consider the design constraints. We confirm the proposed method can solve the optimization problem quickly by experiments.
We discuss the effect of mutation probabilities on the performance of Mutation-Based Evolving Artificial Neural Networks (MBEANN), which is one of the methods of Topology and Weight Evolving Artificial Neural Networks (TWEANN). TWEANN is an approach for evolving both structures and weights of artificial neural networks. TWEANN is expected to perform well than an approach using a fixed-topology neural network with only evolving weight values. The phenotype of MBEANN consists of sub-networks, and the topology of the neural network grows independently within them. Moreover, the structural mutations of MBEANN are designed to reduce the influence on the fitness value. In this study, we focus on the effect of structural mutation probabilities on performance by using a double pole balancing problem without velocity inputs. The performance of MBEANN is compared with NeuroEvolution of Augumenting Topologies (NEAT), which is a typical method of TWEANN. The results show that MBEANN has a higher task achievement rate regardless of the mutation probabilities and task difficulty. From the comparison with NEAT, MBEANN shows a higher performance even with a larger network structure due to the phenotype that consists of sub-networks.