Kohji Tomita

This article introduces a dimension-decomposed geometric learning framework called Sliced Learning for disturbance identification in quadrotor geometric attitude control. Instead of conventional learning-from-states, this framework adopts a learning-from-error strategy by using the Lie-algebraic error representation as the input feature, enabling axiswise space-decomposition (“slicing”) while preserving the SO(3) structure. This is highly consistent with the geometric mechanism of cognitive control observed in neuroscience, where neural systems organize adaptive representations within structured subspaces to enable cognitive flexibility and efficiency. Based on this framework, we develop a lightweight and structurally interpretable Sliced Adaptive-Neuro Mapping module. The high-dimensional mapping for online identification is axially “sliced” into multiple low-dimensional submappings (“slices”), implemented by shallow neural networks and adaptive laws. These neural networks and adaptive laws are updated online via Lyapunov-based adaptation within their respective shared subspaces. To enhance interpretability, we prove exponential convergence despite time-varying disturbances and inertia uncertainties. To the best of the authors' knowledge, sliced learning is among the first frameworks to demonstrate lightweight online neural adaptation at 400 Hz on resource-constrained microcontroller units, such as STM32, with real-world experimental validation.

An emerging application of microfluidic and pneumatic logic in soft robotics is the selective access to tethered pneumatic elements through a reduced number of pneumatic lines (i.e., addressing). In our previous work, we demonstrated two binary addressing systems based on sequences of pressure pulses transmitted in parallel through the same lines later used to actuate the selected elements. This approach supports distributed architectures, parallel actuation, and unlimited addresses, albeit at the cost of longer and complex addressing processes. In this work, we present multi-pressure addressing systems, which solve the primary limitation of binary designs (pressure spikes upon connection of the actuators) and are compatible with multi-pressure components like hysteretic or sequence valves. We also present a strategy for linear networks that maximizes the number of addresses for a given sequence length. We validate our designs with three physical prototypes: a hierarchical module that controls a pneumatic robotic arm, demonstrating the smooth connection of the actuators; a non-hierarchical module with a sequence valve, that triples the network addresses compared to its binary-pressure version (36 addresses, each with 4 simultaneous outputs, using 5 lines); and a linear network with 18 distributed logic modules and air reservoirs emulating adjustable distances, used to evaluate transient pressure drops over extended lengths and to validate the proposed shortened addressing strategy (19 addresses of four or fewer pressure pulses, each with 4 simultaneous outputs, using 5 lines extended up to 1900 m).

In this paper, we present a VR-based control framework for multi-UAV (rotorcraft-type) aerial manipulation that enables simultaneous control of each UAV and its onboard five-degree-of-freedom (5-DoF) manipulator using virtual-reality controllers. Instead of relying on dense button mappings or predefined gestures, the framework maps natural VR-controller motions in real time to vehicle pose and arm joint commands. The UAVs respond smoothly to translational and rotational inputs, while the manipulators accurately replicate dexterous hand motions for precise grasping. Beyond single-platform operation, we extend the framework to cooperative dual-UAV manipulation, leveraging two-hand poses captured via VR controllers to coordinate two UAV-arm systems for payload transportation and obstacle traversal. Simulation experiments demonstrate accurate trajectory tracking and the potential for successful cooperative transport in cluttered environments, indicating the framework’s suitability for telemanipulation, search-and-rescue, and industrial tasks.

The demand for payload transportation has shown exceptional growth owing to its inherent convenience and ability to reduce manpower requirements. To address the limitations of previous studies that primarily employed a single multi-rotor to deploy payloads within a limited reach, we designed a cable-suspended aerial payload deployment system with a large workspace that allowed payload deployment at a distance from the multi-rotor platforms. Our system comprised two multi-rotors, a payload deployment device (PDD, hereafter) with ducted fans, servomotors and a microcomputer. The ducted fans were used not only for swing suppression during payload transportation, but also to ensure that the PDD maintained a certain angle to the vertical plane where the dual multi-rotors were located during the deployment phase. Compared to existing research results, our design has a larger working distance in the vertical and horizontal directions. The controller designed for the PDD was adaptive and robust to variation of the payload mass and distance between the dual multi-rotors. We verified the effectiveness of our design and algorithms through several experiments.

Quadcopter is a type of unmanned aerial vehicle (UAV) that offers excellent flight ability albeit with relatively low payload. Despite its widespread use for transportation, improving its robustness against disturbances and carrying capacity has been challenging. Thus, this study proposed a cable suspended multi-UAVs transportation system with a controller and disturbance observer that is highly robust against disturbances. The primary objective of this study is to design a robust controller and disturbance observer by considering the effect of the cable-suspended payload on the UAV as a part of the disturbance. Consequently, reducing the complexity of the problem and allowing a range of algorithms to be used. A controller and disturbance observer are designed based on backstepping control and the extended state of the model, respectively, to control the flight of the multiple UAVs while carrying a cable-suspended payload. Further, to exploit the agility of quadcopters and avoid obstacles, a virtual structure based leader-follower dynamic formation scheme is introduced to change the formation geometry during the flight with cable suspended payload. This scheme does not make any assumption regarding the status of cable tension during the flight and thus no feedback measurement is required from payload. Moreover, the stability analysis of the proposed system is verified theoretically based on Lyapunov stability and through indoor experiments.

To significantly reduce the occurrence of severe traffic accidents, reducing the number of vehicles in urban areas should be considered. Personal mobility is essential for realizing this reduction, which requires consideration of the last-/first-mile problem. The overall objective of our research is to solve this problem using standing-type personal mobility vehicles as transportation devices; however, to evaluate the feasibility of such vehicles as future mobility devices, it is necessary to evaluate their operation under real-world conditions. Therefore, in this study, experimental and survey data relating to the velocity, stability, safety, and comfort of a standing-type personal mobility device are obtained to evaluate its performance in three different scenarios. The results show that the personal mobility vehicle is socially well received and can be safely operated on sidewalks, irrespective of the gender or age of the driver; moreover, the results suggest that subjects who routinely use a bicycle are adept at avoiding and absorbing the impacts of small holes and bumps, thereby yielding reduced acceleration values (in all directions) and pitch, roll, and yaw rates. This is anticipated to benefit the future development of personal mobility devices and help realize effective and accessible public transport systems, as well as reduce the number of vehicles in urban areas.

This study investigated the effects of supervision on autonomous wheelchairs, which are considered to be a means of future transportation. A rider using an automated wheelchair needs to be vigilant about the failure modes of the system. This study focused on the effects of human factors on an automated system. Several experiments were conducted in public areas to investigate the influence of human factors including acceptability, capability, and usability on the supervision of an automated system by using questionnaires. The average reaction time from the occurrence of system failure until the emergency switch is activated was found to be similar in all the experiments for all test tracks. Regarding the positions of the subject's hand, several patterns for pushing the emergency switch were observed. The method of supervision for pushing the emergency switch was the same among all subjects, even though they had different attitudes. Most subjects had favourable opinions on the automated wheelchair, whereas some found the supervision bothersome. The results reveal the conditions under which future users would prefer to use the automated wheelchair, given the expected cost and functions. The real‐world experimental data are valuable for developing automated wheelchairs and effective human–machine interfaces on automated systems.

This paper presents an automatic convergence estimation method for the reinforcement learning (RL) of autonomous agents. Recently, a multi-agent robot system (MARS) that uses an RL algorithm has been studied in real-world situations. This system can obtain behaviors autonomously through multi-agent reinforcement learning (MARL). However, MARL takes a long time to obtain an optimal or near-optimal solution. Furthermore, the agents continued learning after obtaining of the solution may lead to overfitting. It is difficult for the MARL operator to determine whether the learning has been converged or not, because the operator has to determine the convergence of all learning agents. The convergence of the learning curve indicates that knowledge has been obtained. However, judging the convergence of RL depends on human intuition andexperience, and few studies have discussed convergence estimation methods. Therefore, in this paper, we address development of an automatic convergence estimation method that does not require human judgement. In prior work, we proposed automatic convergence estimation method using fractal dimensional analysis which evaluates the learning curve of the learner as agents, and we confirmed the effectiveness of our previous method by conducting a computer simulation. In this study, we propose a method based on fractal dimensional analysis considering implementation to the robots and evaluate the effectiveness of the method by using an expanded experimental setup that considering a computer simulation and an actual mobile robot environment.