Edwardo Fumihiko Fukushima

Continuum robots (CRs) show great potential in complex environments due to their excellent deformability. For practical applications, the position control of the CRs is an important research field. The common drives of the CRs are rigid motors with mature control schemes. However, the driving force of the rigid motors is often impactive, which may cause safety concerns during interaction. Soft drives based on pneumatic soft actuators (PSAs) can provide compliant driving force for the CRs robot bodies to solve this problem, but it is necessary to comprehensively consider the control of the soft drives and the robot bodies. This article takes a CR with soft drives as the research objective, and proposes a double closed-loop adaptive position control method to achieve the endpoint position control of the CR. This CR includes a length-variable robot body with millimeter-scale diameter and pneumatic soft drives. The kinematic model of the robot body is built based on the piecewise constant curvature (PCC) method, and the static models of the soft drives are derived from the three-element model. Based on these models, we propose a double closed-loop adaptive position control method. The inner loop is used to control the displacements of the soft drives, and the outer loop combines the endpoint position control with the nonsingular fast terminal sliding mode function to adaptively control the endpoint position based on the inner loop. By Lyapunov method, we prove the convergence of the endpoint position error. The effectiveness of the proposed control method is verified through experiments.

Soft robotic grippers offer adaptable and safe manipulation across diverse objects, but they often lack the output force and stability of rigid grippers. Developing fingers that deliver high force while delicately handling small or fragile objects remains a key challenge. Here, we design a humanoid finger with a rigid-flexible-soft structure, composed of rigid tubular bones, flexible joint springs with automatic reset and resistance to deformation under external forces, and a soft silicone skin. A pneumatic membrane actuator is placed between the bones and skin. To test its performance, we build various grippers using the proposed humanoid fingers and conduct experiments such as grasping tofu, pinching 0.1 mm paper, holding a 23.2 cm basketball, and supporting 5.275 kg weights. Our humanoid finger broadens the applications for rigid-flexible-soft robots.

Speech emotion recognition (SER) is pivotal for achieving empathetic and adaptive human–robot interaction (HRI) within Internet of Things (IoT) ecosystems. However, conventional SER methods face computational inefficiency of large self-supervised speech models for HRI system deployment. Moreover, utterance-level emotion labels introduce segment-level ambiguities, degrading model training. To mitigate these limitations, we proposed the multiview logit-based heterogeneous model alignment (MVLogit-HMA) framework, which transfers knowledge from a large self-supervised Trunk (TKG) encoder to a lightweight Package (PKG) encoder through the joint alignment of instance-level contrastive guidance and class-level prototype relations in both the logit space and probability distribution, thereby harmonizing representations across heterogeneous model architectures. Simultaneously, we introduce ambiguity fragment self-weighting (AFSW) to dynamically down-weight unreliable segments and enforce discriminative separation between high- and low-ambiguity groups via an adaptive boundary loss. Comprehensive evaluations on the IEMOCAP and RAVDESS datasets confirm the superiority of our method, achieving weighted accuracy (WA) of 74.17% and 90.02%, and unweighted accuracy (UA) of 74.47% and 90.39%, respectively. Furthermore, a preliminary application in a real-world HRI scenario validates the practical viability of our approach.

Unmanned aerial vehicles (UAVs), such as multicopter drones, are expected to play pivotal roles in varied domains. However, their operation poses challenges owing to the need for stringent safety measures and comprehensive training. This paper introduces a novel mobile robot-restrained UAV trainer to facilitate safe outdoor flight experiments and training sessions, which mitigates the risk of UAV crashes. The proposed system comprises three fundamental components: the UAV (flying object), a coupling-restraint mechanism, and a base platform. The base platform, which operates on the ground, is connected to the UAV via the coupling-restraint mechanism. This configuration allows the base platform to follow the movement of the UAV, thereby enabling training and experimental activities without concerns about potential UAV crashes. This paper outlines the system design, evaluates its impact on UAV motion through simulations, and presents a full-scale prototype for validation.

The advancement of static-dynamic visual emotion recognition plays a pivotal role in the evolution of intelligent and empathetic machines. Currently, the progress of static visual emotion recognition (SVER) faces challenges, primarily the need to balance model efficiency with robust recognition performance. Moreover, obstacles such as noisy data and limited labeled datasets restrict models from effectively learning the appearance features and dynamic dependencies intrinsic to dynamic visual emotion recognition (DVER). In the realm of SVER, facial pixel and semantic representation are derived using a lightweight surface and landmark features embedding network, followed by neuron-energy-based feature fusion-filtering to enhance the fusion of semantic representations positively. For DVER, a Global Anchor-dependent Noisy Emotional Data Filtering (GADNEF) method is employed for noisy label learning, facilitating clip-wise filtering of ambiguous data via iterative computations of frame-wise attention statistics across batches. Furthermore, a self-supervised learning paradigm based on a unified appearance and motion-guided masked autoencoder is implemented, enabling large-scale knowledge transfer tailored for video-based facial expression analysis. Our approach has achieved accuracies of 97.07%, 98.90%, 61.39%, and 92.28% on SVER datasets including KDEF, RaFD, SFEW, and RAF-DB, respectively, while the weighted average recall (WAR) and unweighted average recall (UAR) on DVER datasets such as DFEW and MAFW were 75.12%, 64.37%, 55.26%, and 42.53%, respectively. A preliminary application experiment has also been conducted to validate the practical applicability of our method within human-robot interaction (HRI) scenarios.

The complex nonlinear characteristics of pneumatic soft actuators, such as asymmetric hysteresis, rate-dependence, and mechanical load-dependence, pose a challenge in accurately modeling their dynamics. To address this challenge, this paper proposes a comprehensive dynamic model aimed at describing bidirectional asymmetric hysteresis, rate-dependent, and mechanical load-dependent characteristics of a vertical pneumatic bellows actuator (PBA) system. The dynamic model contains a hysteresis submodel and a load-dependent dynamic submodel. The hysteresis submodel consists of several sets of weighted double-side play (DSP) and weighted dead-zone (DZ) operators connected in series, and it is used to model the bidirectional asymmetric hysteresis of the system. The load-dependent dynamic submodel is built based on the gated recurrent unit (GRU) neural network, and it is used to fit the nonlinear relationship between the displacement of the system and the frequency of the input air pressure as well as the mechanical load. The model parameters of the hysteresis submodel and the load- dependent dynamic submodel are determined by intelligent optimization method and neural network training method, reseparately. The fitness value (FV) between the output of the dynamic model and the experimental data is calculated to be 96.1736%, demonstrating that the parameters of the dynamic model are valid. We conduct six set of experiments to compare the model output with the experimental data, and calculate the root-mean-square errors and the maximum error, respectively. The experimental results show that, the root-mean-square error remains consistently below 2.7700%, while the maximum error remains below 8.4000% across all experiments, thereby substantiating the validity and generality of the proposed model.

Compared with conventional rigid-link robots, bionic continuum robots (CRs) show great potential in unstructured environments because of their adaptivity and continuous deformation ability. However, designing a CR to achieve miniaturization, variable length and compliant driving force remains a challenge. Here, inspired by the earthworm in nature, we report a length-variable bionic CR with millimeter-scale diameter and compliant driving force. The CR consists of two main components: the robot body and soft drives. The robot body is only 6 mm in diameter, and is composed of a backbone and transmission devices. The backbone is divided into three segments, and each segment is capable of adjusting its length and bending like the earthworm. The maximum length variation of the backbone can reach an astonishing 70 mm with a backbone's initial length of 150 mm, and the maximum bending angle of each segment can reach 120 degrees. In addition, we develop soft drives using pneumatic soft actuators (PSAs) as a replacement for the rigid motors typically used in conventional CRs. These soft drives control the motions of the transmission devices, enabling length variation and bending of the backbone. By utilizing these soft drives, we ensure that the robot body has a compliant driving force, which addresses users' concerns about human safety during interactions. In practical applications, we prove that this CR can perform delicate manipulations by successfully completing writing tasks. Additionally, we show its application value for detections and medical treatments by entering the narrow tube and the oral.

Soft manipulators have been widely developed due to their manipulation safety, design versatility, and motion dexterity. However, accurate kinematic modeling and control of soft manipulators are still challenging due to the complex nonlinear hysteresis in their motion characteristics. This article presents the fabrication of an elephant-trunk-inspired soft manipulator (ESM) using pneumatic bellows actuators (PBAs) and proposes a kinematic modeling and control method considering hysteresis. The ESM contains three soft arms and a pneumatic claw, while each soft arm consists of four PBAs connected in parallel. The forward kinematics of the ESM are modeled using the piecewise constant curvature (PCC) method. The kinematic model of the ESM consists of three parts: actuator space, configuration space, and task space. To address the inability of the PCC method to characterize the nonlinearity of soft materials, the hysteresis of the PBA in the actuator space is modeled using the Prandtl–Ishlinskii model. The inverse kinematic model of the ESM is derived from its forward kinematic model, and on this basis, a feedforward controller is developed to achieve position control of the ESM. The positional workspace of the ESM is analyzed using the Monte Carlo method. The validity of the kinematic model is verified experimentally. Position control and application experiments are conducted to demonstrate the practical application of the ESM.