A training course for performance evaluation of infrastructure inspection aquatic robots based on the Performance Evaluation Procedures issued by NEDO/METI was held in JFY2018 – 2020, and 41 trainees attended. The document consists of ‘specific ability test’ and ‘mission-type test’, and the latter is a more practical test based on the actual dam inspection scenario. In the course, trainees learned concepts and theories of the performance evaluation of robots and conducted performance evaluations of ROVs by referring to the test procedures. The evaluation reports submitted by the trainees exhibited wide variation in terms of evaluation items and criteria. This infers that the interpretation of the Procedure document varied amongst trainees, which was derived from what kinds of performances each trainee seemed to prioritise. The analyses of the evaluation results have provided clues for practical application and further improvements of the mission-type test of the underwater inspection robots.

Imitation learning is a promising paradigm for training robot agents; however, standard approaches typically require substantial data acquisition – via numerous demonstrations or random exploration – to ensure reliable performance. Although exploration reduces human effort, it lacks safety guarantees and often results in frequent collisions – particularly in clearance-limited tasks (e.g. peg-in-hole) – thereby, necessitating manual environmental resets and imposing additional human burden. This study proposes Self-Augmented Robot Trajectory (SART), a framework that enables policy learning from a single human demonstration, while safely expanding the dataset through autonomous augmentation. SART consists of two stages: (1) human teaching only once, where a single demonstration is provided and precision boundaries – represented as spheres around key waypoints – are annotated, followed by one environment reset; (2) robot self-augmentation, where the robot generates diverse, collision-free trajectories within these boundaries and reconnects to the original demonstration. This design improves the data collection efficiency by minimizing human effort while ensuring safety. Extensive evaluations in simulation and real-world manipulation tasks show that SART achieves substantially higher success rates than policies trained solely on human-collected demonstrations. Video results available at https://sites.google.com/view/sart-il.

This paper presents a control design framework for collaborative object transportation using differential-drive robots. Specifically, pairs of robots are required to navigate toward a target pose while maintaining predefined inter-robot distances and relative orientations, as well as avoiding collisions with unknown static obstacles and other robot groups. To achieve this, both centralized and distributed quadratic programming (QP)-based controllers are developed, where control barrier functions (CBFs) are incorporated as constraints to handle multiple control objectives simultaneously. The performance of the proposed controllers is evaluated in both simulations and real-world experiment through various scenarios, including navigation in complex, office-like environments.

Model Predictive Control (MPC) is widely adopted across robotic platforms, yet optimal performance hinges on careful tuning of the quadratic cost, where weighting matrices Q and R balance state tracking and control effort. This study introduces a Convolutional Weighting-Enhanced MPC (CMPC) that dynamically adjusts cost weights based on reference deviation by convolving the reference signal with a predefined shaping kernel. The framework applies to standard MPC as well as Nonlinear MPC (NMPC) and Adaptive MPC (AMPC) and related variants. Because the convolution updates Q and R without adding decision variables or a separate optimization layer for weight adaptation, computational overhead is significantly reduced relative to adaptive schemes. Beyond formulation, we rigorously analyze stability and robustness, including CMPC behavior under noise, and we derive sufficient conditions guaranteeing preservation of closed-loop properties. The approach is validated in a quadrotor simulation with two demanding trajectories: one using time-optimal path parameterization and the other employing minimum-snap planning, chosen to stress high-performance motion requirements. We benchmark CMPC against Linear MPC, NMPC, AMPC, and a Linear Quadratic Regulator (LQR). Across scenarios, the convolutional framework consistently yields lower tracking error and reduced energy consumption compared to conventional methods, supporting its suitability for energy-efficient next-generation UAV reference tracking.

Human action anticipation remains a key challenge to achieve efficient human-robot interaction due to the difficulties to learn the higher level of abstraction. This work explores three action anticipation pipelines as a guideline for future work. Specifically, two pipelines adopt a top-down approach: they recognize current actions and then anticipate future actions using either traditional machine learning models or Large Language Models (LLMs). The third pipeline follows a bottom-up strategy by first forecasting future motions and then inferring actions. Our results show that top-down pipelines achieve higher accuracy and robustness, demonstrating the advantage of abstract reasoning over direct motion-based inference.

This paper presents an event-triggered fixed-time sliding mode control scheme for trajectory tracking of Euler–Lagrange (EL) systems under matched disturbances. A nonlinear sliding manifold to guarantee fixed-time reaching and sliding phases. In order to ensure that the control is free from singularity, a weighted sampling error with a switched triggering mechanism is designed. Moreover, the proposed triggering law exhibits a strictly positive inter-event interval globally. Using Lyapunov stability analysis, we demonstrate the existence of a practical sliding mode band (PSMB) in a fixed time. Using the results of fixed time practical sliding mode, we further show the existence of fixed time practical stability for the closed-loop dynamics. Moreover, we explicitly show that the PSMB and inter-event time can be tuned by the event design parameter with a trade-off. Numerical simulation of a two-link manipulator and experimental results obtained through real-time validation on a single-link manipulator validate the proposed theory. The controller achieves convergence within the calculated fixed-time bound, with reduced control updates relative to periodic SMC, and maintains robust tracking under unmodeled friction.

Modular robots, which can form various configurations by connecting multiple modules, have the ability to perform diverse tasks by reconfiguring themselves. However, due to the torque limitations of a single module, lifting multiple modules simultaneously is challenging, making it difficult to construct manipulators for manipulation tasks. In this study, we propose TYCOON (modular robot capable of Torque sYnthesized COllabOratioN), a modular robot that can synthesize more torque by transmitting torque between modules. In conventional modular robots, each module functions as a joint when assembled (joint-drive function), enabling them to lift or rotate other modules. TYCOON modules have an additional torque-transmission function that enables modules to transmit their torque to adjacent modules, allowing the torques of multiple modules to be synthesized and increasing the driving torque per joint. To realize this, TYCOON employs a Complementary Intermittent Gear Mechanism consisting of two gears with complementary teeth, ensuring that when one gear is engaged the other remains disengaged, achieving two functionalities with a single drive mechanism. Evaluation experiments showed that while a single module could lift only two modules, TYCOON could lift up to four connected modules. This capability enabled a 6-DOF serial-link manipulator, demonstrating the effectiveness of torque transmission in modular robots.

Soft actuators, which are the central technological element of soft robots, are expected to possess characteristics similar to those of biological muscles. Therefore, we proposed an artificial muscle using DME combustion (hereinafter referred to as ‘combustion artificial muscle’). Combustion artificial muscles have lower inertia, faster response, and can generate greater force than motor or pneumatic artificial muscles. However, the reliability of existing models, verification of scale effects, and quantitative evaluation of power are insufficient. Therefore, this study examined models in isometric contraction force and displacement response experiments, quantitatively evaluated power, and verified scale effects. Although the square-cube law limited the performance improvement of general fluid actuators by increasing their size, the performance of the combustion artificial muscle improved as its diameter increased. The larger diameter combustion artificial muscles exhibited higher force, displacement, and power while maintaining responsiveness. Mathematical models were developed and validated, providing insights for further improvement. These results verify the advantages of scaling combustion artificial muscles and highlight the potential for high performance applications in soft robotics.

This paper presents a motion analysis framework for an athlete wearing sport-specific flexible prosthesis based on the soft-rigid hybrid-link system. Such a motion analysis is a challenging problem because we need to consider the interaction force between the rigid human skeleton system and a flexible prosthesis. However, most of human musculoskeletal models are based on the computation framework of a rigid-body multi-link system. Recently in soft robotics research field, fast and efficient modeling methods were developed for a flexible rod deformation, which allows us to build a hybrid-link system that integrates rigid link and soft bodies in a unified formulation. We apply inverse kinematics of the hybrid-link system to motion reconstruction from a motion captured data, and also present the estimation of the joint torques and ground reaction force by inverse dynamics. Through a human subject experiment, we show that the inverse dynamics achieved approximately 12% error on the ground reaction force estimation. Furthermore, we provide the muscle force estimation considering muscle amputation and interaction force with the prosthesis leg deformation.

Robotic systems for capturing images for remote visual inspections of train bogies – in a manner that reduces labor and time – are being developed. One method for intuitively determining the robot’s camera pose is to use the inspector’s head pose recorded by a head-mounted display (HMD). However, a method for extracting the camera pose from which the robot can image the inspection points from numerous head poses has not yet been investigated. Assuming that the inspector’s gaze points during inspection are highly concentrated, we propose a method that extracts the camera pose aligned with the dense clusters of gaze points collected by the HMD and a robotic system that assumes the extracted camera pose to capture images. It was experimentally validated that inspection-related gaze points are more clustered and the camera pose for capturing images can be extracted from the clustering results. The robotic system was tested in a real-world setting in which it captured images at 11 inspection points during a railway company’s periodic inspection of a train bogie. The robot-captured images were verified as usable for inspection by a human inspector, and that verification confirms the practicality of the robotic system.