In general, workers use a hose connected to a vacuum car to clean up iron ore that has fallen from the conveyor and accumulated. However, vacuuming by workers requires the conveyor to be stopped during the operation to prevent contact between workers and the conveyor, which leads to a decrease in productivity. Therefore, it is necessary to develop a vacuum work robot to replace the worker. Additionally, using a remote-controlled vacuum work robot to clean the pile makes it difficult to vacuum the sediment because the nozzle moves away from the sediment when the robot rides up on sediment. Hence, it is important to automatically control the nozzle position to maintain the proper distance between the nozzle and sediment. In this study, we purposed to develop a robot efficient vacuuming operation that can automatically control the nozzle position and reduce the amount of leftover suction. Therefore, we proposed a nozzle position control method that enables the robot to vacuum sediment even when it rides up on sediment. In conclusion, we proved the effectiveness of the proposed method through experiments with automatic nozzle position control on a test course simulating the work environment.

This paper proposes a multi-robot cooperative localization method based on the reliability evaluation of positioning and complementary information sharing, with the aim of contributing to the practical application of multi-robot systems. In conventional methods, robots are often separated into two groups, parents that serve as a reference for positioning and children that depend on its information, but errors in the parent robot can spread to the entire robot system. Therefore, we propose a method to estimate robots' positions by dynamically switching reference robots without specifying a particular robot as a parent. We evaluate the reliability of positioning based on the number of positioning references observed by each robot and the uncertainty of each robot's position estimation. After that, robots complementarily share information and switch the reference robot based on these references. The effectiveness of the method was verified using data obtained from actual robots. The results show that the proposed method can provide stable position estimation for the entire robot swarm better than the case where the parent robots are predetermined, even if there were robots with position estimation errors.

This letter presents a formation flight method for multirotor unmanned aerial vehicles using hierarchical optimal control (HOC). HOC enables to tune trade off relationships between individual and cooperative targets existing in formation flight via simple weight parameters tuning. The aim of this letter is implementing the formation flight method for actual quadcopters and discussing its possibility and effectiveness for real applications through demonstrations. Demonstrations show that relative state errors between quadcopters can be accurately controlled by HOC more than general linear quadratic regulator. The software used in this work is publicly available.

For autonomous navigation, mobile robots are required to avoid collisions with dynamic obstacles such as pedestrians. In a previous works, we have proposed a motion planner based on CNN with RGB and depth image inputs. In order for a robot to plan avoidance motions taking the moving direction of an obstacle into account, we propose a motion planner using optical flow images as the inputs. Through autonomous navigation, we show that the robot based on the motion planner is able to avoid not only dynamic obstacles, but also static obstacles.

A novel interface we have developed allows the user to operate the robot intuitively because the user operates it according to a Cartesian coordinate system. However, interferences between axes in this interface cause unintended input by the user, reducing operability. In this study, we examine whether machine learning models can identify intentional or unintentional input in this interface. Input values during interface operation are acquired, and a model is built for each axis to estimate the input state. Models for all axes achieved an F1 score of 0.97 or higher.

The optimal grasping points of an object with a robotic hand depend not only on external wrenches but also on the position of the object's center of gravity (COG). Depending on object's type, the COG position of an object may be indeterminate. In this paper, introducing a candidate COG position set as well as a required external wrench set, we propose a method to derive the optimal grasping points for an object with indeterminate COG. We also confirm that the proposed method works properly with numerical examples. In addition, we examine the improvement of efficiency of the optimization method.

In this paper, we propose a novel model for multi-agent symbol emergence that integrates Gaussian Process Latent Variable Model (GPLVM) and neural networks, in which shared symbols between two agents can emerge. In the proposed model, the agents create symbols bottom-up by interacting with the environment and share their meanings by interacting with others. Using GPLVM, the symbols are represented as continuous variables that are more expressive than discrete variables. For the inference of symbols in GPLVM, we utilize Metropolis-Hastings Naming Game (MHNG). MHNG is a method that enables agents to acquire shared symbols and their meanings by communicating the symbols between them without directly observing the other's internal states. Furthermore, we introduce the neural network called Neural Conversion Adapter (NCA) that converts the features extracted from observation to low dimensional latent variables that are internal states of each agent. NCA facilitates the inference of appropriate latent variables for representing symbols. Experiments show that the GPLVM-based symbol emergence model can generate shared symbols and that more explicit symbols can be learned by combining NCA with GPLVM.

Mobile robots will play a key role in future extraterrestrial construction on the lunar surface. As the mobile robots traverse around the lunar regolith, their running gears, such as wheels, often generate ruts on the terrain. Those ruts after several wheel passes may significantly change the robotic mobility, which is called a multi-pass effect. While the multi-pass effect for heavy vehicles was addressed in the past, this paper aims to analyze the multi-pass effect for small-lightweight wheels experimentally. The experimental data obtained from our wheel test bench were qualitatively consistent with the past research on the multi-pass effect on large wheels. However, we found that the wheel slip ratio varied less than a few percent even though the number of runs increased. These results are then discussed based on the interaction mechanics of the wheel on loose sand.