In recent years, there has been a paradigm shift in robotics and autonomous systems, with the transition from preprogrammed, structured environments to highly dynamic and unstructured real-world settings. During this transition, physical interaction has become increasingly crucial, putting tactile and proximity sensing at the forefront of research. Accurately and continuously capturing the state of an environment from “approach” to “contact” is indispensable for safe human-robot collaboration, handling unknown or deformable objects, and achieving dexterous manipulation.

This special issue focuses on the latest advancements and future prospects in the fields of tactile and proximity sensors. This collection contains 15 excellent papers that broadly include three major research trends.

The first involves the fundamental development of novel sensing hardware. These studies presented ambitious hardware innovations based on new materials and principles, such as flexible microstrain sensors, catheter-type sensors for medical applications, microelectromechanical systems (MEMS)-based multimodal devices, and lightweight optical sensors constructed using transparent flexible resins.

The second trend highlights the practical applications of sensor integration and advanced data processing. Combining algorithms and multimodal methods, such as seamless hybrid sensing using Time-of-Flight (ToF) and capacitance, computational approaches for estimating the internal stiffness of flexible objects, and machine learning for predicting grasp configurations, has demonstrated ways to overcome the limitations of individual sensors to acquire more robust perception.

Furthermore, the third trend shows application development for real-world tasks. This includes manipulation control for continuum robots and highly dynamic tasks, safety evaluations in human-robot interactions, wearable haptic feedback devices, and precise surface inspection systems for industrial applications.

The editors hope that all readers will consider the selected papers informative and inspiring. We hope that this collection will act as a catalyst for expanding this vital research area and promote further interdisciplinary contributions. Finally, we express our deepest gratitude to the authors for their outstanding contributions and the reviewers for their dedicated time and insightful feedback.

To enable quantitative palpation in vivo, we previously developed a catheter-type tactile sensor that uses a polyvinylidene fluoride (PVDF) film for detecting lesions based on surface changes. This study investigates the effects of the structural parameters for the sensor on the piezoelectric output. A coupled electrical-structural finite element analysis (FEA) is used to simulate the displacement of a sensor tip composed of silicone rubber layers, a PVDF film, and a plastic substrate film. The FEA results indicated that encapsulating the plastic film in rubber increased the output charge by approximately 56.5%, primarily due to enhanced strain caused by lateral expansion of the silicone rubber. Increasing the plastic film thickness and the distance between the neutral plane and the PVDF film was also found to increase the output charge. In addition, the sensor output was larger when the lower rubber layer was made thicker than the upper rubber layer. However, the material used for the substrate film was predicted to have the dominant effect on the sensor output, with the largest output being achieved for a film with a large Young’s modulus. To confirm the FEA results, we fabricated prototype sensors that used plastic, steel, and titanium films, and experimentally evaluated their performance. The results indicated that an appropriate choice of film material could increase the sensor output by 18.9 fold.

This study proposes and develops a surface-scanning tactile sensor and mapping system that amplifies strain gauge signals to enable the detection of minute defects. The research objectives include: (1) developing high-strength, high-sensitivity sensing for distortions of tens of micrometers; (2) constructing a system capable of rapidly scanning a large inspection area; and (3) visualizing and digitizing defect locations. To achieve these objectives, three key approaches were implemented. First, we developed a tactile sensor with a simple, flexible structural design. By arranging pins parallel to the strain gauges, the sensor effectively amplifies mechanical deformation to increase output. Experiments showed that the proposed sensor achieves signal amplification of approximately 7.6 times compared to a strain gauge-only sensor and produces an output approximately 3.3 times greater than that of a conventional sensor. Second, because the sensor cannot independently determine positional information, we estimated positions using a marker-based approach. Specifically, we measured the distance between a target marker attached to the sensor and a reference marker during scanning. The accuracy was evaluated using a dimensionless error metric normalized by the 60 mm gauge length, and the system yielded an error of 7.6% with a standard deviation of 5.5 mm. Third, we developed a system capable of probabilistically estimating the precise location of micro-surface distortions. The strain gauges used in current tactile sensors have a gauge length of 60 mm, which introduces the limitation that the exact location of a micro-surface distortion cannot be determined from a single measurement. To address this limitation, we developed a method that increases the probability of detecting micro-surface distortions through repeated scanning, thereby progressively narrowing the estimated detection range.

This study aims to enhance tactile sensing for practical robotic applications by enabling the acquisition of dynamic contact information in optical tactile sensors. Conventional optical tactile sensors measure displacement and torque with high precision by detecting the deformation of transparent flexible resin using photoreflectors; however, they do not fully exploit information from minute vibrations or dynamic contact events. In this work, we propose a lightweight, low-cost, and robust optical tactile sensor capable of texture recognition and slippage detection without relying on acceleration sensors or piezoelectric elements, offering a simpler and more durable alternative to conventional high-definition camera-based approaches.

In this study, a MEMS tactile sensor was designed and fabricated, with the aim of achieving integrated measurement of object contact, thermal, and proximity sensation. This sensor integrates a Si photoresistor for light detection, a microcantilever with a NiCr strain gauge for force detection, and an Au resistance thermometer detector with a heater for temperature detection. Furthermore, since both the strain gauge and the photoresistor exhibit temperature dependency, temperature compensation was applied to the measured data. Thus, it was demonstrated that a single sensor can dynamically and simultaneously measure different modalities: object contact force, thermal, and proximity sensation. Sensors equipped with both tactile and proximity capabilities are expected to enable accurate manipulation and control of grasping force.

In recent years, collaborative robots (cobots) that can operate safely with humans have gained popularity. Proximity and tactile sensors contribute to the safe operation of cobots in shared workspaces. This study proposes a time-of-flight (ToF) sensor and self-capacitance proximity and contact sensor that combines five wide field-of-view multi-zone ToF sensors with a self-capacitance electrode, allowing for installation on curved surfaces such as non-driven regions near joints and achieving wide-range measurements from proximity to contact. The combination of ToF and self-capacitance sensing allows seamless detection from non-contact to contact with fewer blind spots. Furthermore, real-time control based on the acquired data was implemented, demonstrating improved safety and operational efficiency during collaborative tasks by reducing robot speed and stopping its motion when proximity is detected. The proposed method allows for both curved-surface installation, including non-driven regions near joints, and continuous detection of proximity to contact in cobot safety systems.

In this study, we defined a proximity sensor based on the fusion of an optical reflection intensity sensor and optical time-of-flight (ToF) sensor as a wide-range precise proximity sensor (WrPPS). This sensor can detect objects over a wide range, has low dependence on the physical properties of the detected object, and can be configured in a compact form. To allow the broad application of this sensor, we propose a WrPPS Single Board that packages the minimal configuration of this sensor onto a compact printed circuit board, allowing applications wherever this board can be mounted. Furthermore, to improve the accessibility of this board and allow easy application by anyone, we sell the board at a low price and release the fusion software for the intensity and ToF sensors as open-source software. The distance measurement accuracy of this sensor was quantitatively evaluated through experiments involving variations in the reflective properties, shape, and pose of the measurement target. To demonstrate the wide applicability of this sensor, we present examples of its applications, including grasping of compliant objects, tactile sensing in a stuffed robot, slip detection during walking, and agricultural plant sensing.

A method is proposed to support robotic regrasping by leveraging tactile data to extrapolate unseen contact regions. Initial tactile feedback from multimodal capacitive sensors mounted on robotic fingers was used to classify the object shape into prototypical categories. Based on this classification, shape-specific extrapolation strategies extend the tactile map beyond the initial contact area, providing a computationally efficient estimate of potential contact without requiring complex physical simulations. The extrapolated regions were evaluated against measured contact data collected via a systematic grid-based scan using three metrics: the tactile centroid deviation, defined as the Euclidean distance between the geometric centers of binary contact regions; the grasp success rate estimated by a pretrained grasp assessment network; and the structural similarity index to assess local structural fidelity. Experiments on cuboidal, spherical, and cylindrical objects demonstrated the effectiveness of the approach in predicting unseen contact and identifying safe zones where extrapolated and real contacts align. The results showed reliable performance for small rigid objects, with the shape classifier achieving 94.3% accuracy on a held-out test set. However, a reduced accuracy was observed for larger, highly curved geometries, likely due to the limited curvature resolution of the tactile sensors. Large-diameter cylinders are occasionally misclassified as cuboids. Potential improvements include enlarging the dataset, refining the classifier, and integrating high-resolution sensors to enhance adaptability and precision.

To achieve global palpation efficiently and without the use of tactile sensors, this study proposes mechanical ghost palpation (MGP), which applies the concept of ghost imaging (GI) to palpation, and verifies its basic principle. However, unlike general GI, which uses light as a medium, the configuration of the pressing patterns in MGP, which involves mechanical contact, presents system-specific challenges: from the correlation between the pattern structure and pressing depth, to the replacement and mass production of the patterns. To overcome these issues, in this work, we proposed a pattern configuration method that imposes constraints on the number of contact points and the centroid of their positions, and further proposed a technique to virtually increase pattern diversity by using a rotation operation on two square patterns. In the experiment, we estimated the stiffness map of a silicone gel phantom containing an embedded inclusion within a 5×5 grid by performing a total of 8 times pressing operations (using two pressing patterns). The results confirmed that the inclusion’s location was included within the top three highest-stiffness positions in the MGP-estimated stiffness map. This achievement demonstrates the potential of MGP to efficiently perform global palpation independent of tactile sensors.

This paper proposes a surface inspection method using a double-roller tactile image sensor. The roller-type tactile image sensor provides high spatial resolution by employing a camera and enables continuous contact with the object surface via the rollers. However, a single sensor can acquire a contact image from only one side of the object. To address this limitation, we developed a double-roller tactile image sensor capable of capturing contact images from both sides by sandwiching the object between two rollers. The relationship between applied force and image response was investigated through sensitivity experiments using a force gauge and nylon thread. Experiments using models simulating food and foreign objects revealed that a sufficiently large response was obtained with a pressing force of 0.1 N even for small objects with a diameter of 0.25 mm, while differences in material hardness and foreign object size resulted in differences in the contact image. Furthermore, we performed an experiment using shrimp as an example of food inspection and confirmed that the proposed sensor can successfully capture images of shell fragments remaining on the shrimp.

This paper investigates state estimation for continuum robotic fingers in feature-sparse and dynamic in-hand manipulation environments. Continuum fingers, inspired by continuum robots, offer enhanced flexibility and wider reachable workspace compared with conventional rigid-link fingers to enable grasping and manipulation tasks. However, they lack encoder-based joint angle measurements, making it difficult to determine fingertip positions, particularly under external forces during contact. This limitation hinders precision grasping and prevents the full exploitation of their high dexterity. To address this challenge, we developed a simultaneous localization and mapping framework for continuum fingers using proximity sensors. Unlike conventional simultaneous localization and mapping that assumes feature-rich environments, grasping scenarios present feature-sparse conditions with limited environmental information. We propose an estimator that fuses proximity sensing with a constant-curvature kinematic prior by replacing encoder angles with virtual joint angles. The key idea is to leverage the designed in-hand elements, namely opposing fingers and the palm, as stable reference geometry. Simulations demonstrate that the proposed estimator outperforms a kinematics-only baseline by suppressing bias and reducing position error. Three-dimensional contoured palms enhance observability, with a composite wavy palm yielding the smallest errors without temporal drift. These findings indicate that the designed in-hand geometry combined with temporal map management enables effective state estimation for continuum fingers in feature-sparse and dynamic grasping scenarios.

High-speed sensors, such as high-speed cameras and optical proximity sensors, enable the detailed temporal measurements of physical phenomena that exceed the dynamic capabilities of conventional industrial robots. However, effectively leveraging this sensor information for robot motion planning remains challenging because of the temporal-scale gap between sensors and robots. This paper proposes a motion planning method that extracts the task-relevant meta-information of target phenomena from high-speed sensor data and generates feasible trajectories by considering robot constraints. The information extraction process identifies task-relevant characteristics from high-speed sensor data. To integrate heterogeneous sensor information and enable trajectory adaptation, we employed multiple virtual-dynamics-based control (MVDC), which can asynchronously integrate heterogeneous sensors with different measurement principles. To validate the proposed method, we conducted a case study in which a conventional industrial manipulator grasped a pendulum at its equilibrium point, the most challenging position. The system integrated global measurements from a 1 kHz high-speed camera with local measurements from proximity sensors using MVDC to predict the pendulum period and optimal grasping timing. Experimental results demonstrated that the proposed method enables successful grasping by bridging the temporal-scale gap between high-speed sensors and conventional robots through information integration.

With the development of the internet and the availability of sufficient bandwidth, the performance of haptic devices is considered important because device specifications directly determine the comfort of wearing and the tactile sensations that can be transmitted. The Funabot-Grab is a haptic device that evokes a grabbed tactile sensation. Funabot-Grab can exert a tightening force by controlling the air pressure applied to the artificial muscles. There are seven artificial muscles on the fabric, and the tightening force of each artificial muscle is controlled independently. In a previous study, we used a model arm to incorporate Funabot-Grab into a tactile transmission system with the aim of using it as a tactile internet device. We conducted a preliminary experiment with three subjects as a precursor to the full-scale experiment. It was found that it is possible to transmit the static contact force and its position, even to the subjects. However, it became clear that further improvements to the device are needed for it to be worn optimally on every human body, considering the large degree of individual variation, and to evoke a smoother, continuous tactile sensation.

Wearable assistive robots have been studied widely to reduce the physical burden of tasks such as lifting and caregiving. Most existing systems, however, control only the joint torque. Even though wearable robots transmit actuator forces through direct contact with the body, these systems do not focus on managing the contact force generated at the human–robot interface. Our previous study developed wearable assistive robots capable of measuring and controlling the contact-force distribution. We verified the feasibility of evaluating safety and control performance using a single-joint arm-mounted robot and a multi-joint whole-body robot. However, the influence of force distribution on assist performance was not been sufficiently examined for multi-joint robots under assist conditions. The current study aimed to experimentally verify the effectiveness of using contact-force distribution information in a torso-mounted multi-joint robot (through experiments involving seven participants). The results showed that whereas the actual motor torque tracked the commanded torque under conventional torque-based control, the contact-force distribution experienced by the wearer varied across participants. This finding indicates that for wearable robots with complex joint structures, assessments of wearing states and assist effectiveness require an evaluation of the contact-force distribution in addition to the torque.

This study proposes an object identification method that combines high-precision tracing motion by an industrial robot with vibration information obtained from a polyvinylidene fluoride (PVDF) film sensor mounted on the finger surface. In conventional tactile sensing technology, methods that place sensors on the contact surface face challenges such as reduced durability due to wear and limited design freedom for the finger surface. The proposed method overcomes these issues by exploiting vibration propagation to physically isolate the sensor from the contact point. In this paper, we first investigate how tracing speed and pressing depth affect the vibration spectrum based on the mechanism of vibration generation. We then verify, through fundamental experiments using sandpaper, how finger material influences vibration transmission characteristics. The results demonstrate that a lightweight resin finger is advantageous for acquiring high-frequency vibrations and that strict control of tracing speed is essential for ensuring identification accuracy. Based on these findings, we conducted identification experiments using wide-band vibration information from 0 Hz to 8,000 Hz as features. The experiments targeted 33 types of diverse objects selected from real-world environments according to onomatopoeic classification. Evaluation with a support vector machine achieved an extremely high identification rate exceeding 99% for both finely textured sandpaper and real-world objects under high-precision robot control. Furthermore, high identification rates were maintained even when the tracing position and pressing depth were varied randomly, demonstrating the effectiveness and robustness of the proposed method.

This study proposes a human tracking algorithm that combines vision-based object detection and body heat tracking. The small autonomous mobile robot developed in this study is equipped with an ultracompact camera and three thermal array sensors. We propose a hybrid human tracking method based on camera images and body heat information. In particular, the proposed machine-learning-based image recognition method employs Faster Objects, More Objects, an object detection algorithm optimized for resource-constrained edge devices such as microcontrollers. This algorithm generates an inference model focusing on the clothing features of the target person, which is implemented in the robot control system. Furthermore, a human-robot distance estimation model is derived to calculate the tracking distance between the target person and the robot using a monocular camera. Experiments demonstrated that the proposed inference model enables robust tracking even when the target is partially occluded by another person. Additionally, the body-heat-distribution-based tracking method employs the summation of multiple thermal pixels to improve tracking performance. In this study, tracking performance was improved by emphasizing the target region in the thermal image using four-pixel, nine-pixel, and one-column pixel summation methods. The experimental results demonstrate that the four-pixel summation method, which minimizes fluctuations in the robot posture angle during motion, is the most suitable for the proposed tracking algorithm. The proposed algorithm is effective for hybrid control in environments where illumination conditions change abruptly between bright and dark areas.

Teams participating in robot competitions commonly use LiDAR-SLAM for robot navigation. However, when there are significant differences between the pre-created point cloud map of the environment and the point cloud data obtained during autonomous driving, or in open environments where acquiring point cloud data is difficult, the robot may lose its self-position and often fail in autonomous driving. Humans can reach their destinations even in unfamiliar environments by relying primarily on visual cues along with guidance information such as maps and signs. It can be said that humans rely on visual information when navigating. Similarly, if robots navigate using visual information like humans, it may become possible to achieve autonomous driving without requiring pre-acquired dense point cloud maps. In this paper, we investigated how humans walk and what they focus on while walking, both in real world and when remotely controlling a robot using VR (including cases where surrounding information other than the road was removed). The results indicated that if the recognition of the road is possible, it may be feasible to complete a course without a pre-existing map. Based on these findings, we developed a simple navigation system using road recognition using vanishing points. Its effectiveness was confirmed through driving experiments conducted on a university campus.

We developed a baseball pitching practice ball to provide the real-time visual feedback of the spin rate by changing the surface color according to the rotational speed of the pitch. This system eliminated the need for external display devices such as tablets or monitors, thereby reducing the system complexity and cost. Furthermore, it allowed players, coaches, and spectators to simultaneously visually perceive the spin rate, facilitating immediate and intuitive feedback during training.

Margin of stability (MoS) is a metric used to assess an individual’s dynamic postural stability during walking. It can identify those at risk of falling and enhance their awareness of preventive measures. Although accurately computing the MoS requires capturing the motion of the entire body, previous research has shown that the kinematic information, specifically the three-axial translational velocities of a single bony feature, can estimate the MoS value to some extent. Such information can be obtained from an inertial measurement unit installed in portable devices such as smartphones and smartwatches. Nowadays, it is common for individuals to have two or more such devices. The primary objective of this study is to determine which combination of data from two or three bony features can most accurately predict the MoS. We used a camera-based kinematic database of healthy Japanese walkers, selecting gait data from 30 male and 30 female participants aged 60 and above. The analysis method involved principal motion analysis, a linear predictive model for multi-dimensional time-series data, to predict the MoS, and used cross-validation for stable model assessment. The results exhibited that combining two bony features generally outperform single features, and the combination of the sacral crest and T10 vertebra was the most effective in predicting MoS with an RMSE of 0.0070 m, followed by the combination of the sacral crest and right toe. We did not find any evidence that the combination of three bony features would result in substantially better accuracy than that of two features. This finding suggests that if IMU-equipped devices are placed on these two body parts, they could better assess their risk of falling.

In recent years, the Seto Inland Sea has experienced extensive blooms of moon jellyfish (Aurelia aurita), resulting in substantial disruptions to the local fishing industry. Large aggregations of jellyfish clog fishing nets, damage gear, and reduce fish catches, causing significant economic losses for coastal communities. In 2023, unusually high jellyfish densities were recorded from April onward, and fishing operations in some areas were either severely curtailed or temporarily suspended due to the intensity of the blooms. Despite ongoing monitoring and localized mitigation efforts, no effective large-scale removal methods currently exist that can be rapidly and efficiently deployed under real marine conditions. To address this pressing issue, we developed a remotely operated vehicle (ROV) equipped with a custom-designed suction-and-shredding jellyfish removal device, capable of capturing and removing jellyfish in situ. The ROV was designed with a dry weight of less than 40 kg, enabling two operators to deploy and retrieve it without specialized lifting equipment. Field experiments conducted in the Seto Inland Sea demonstrated that, once a jellyfish was detected by the operator, removal typically required approximately 20 s. Additionally, the fragments collected after removal measured less than approximately 50 mm in diameter. According to previous studies on jellyfish regeneration, fragments of this size are incapable of regenerating. These field trials provided valuable insights into the ROV’s handling, effectiveness, and practical applicability. The results suggest that the proposed method has the potential to substantially reduce fishermen’s workload, mitigate economic losses, and contribute to sustainable marine resource management by providing a rapid and efficient response to large-scale jellyfish blooms.

We have been working on a system that alerts visually impaired people who get closer to fall risk areas, such as platform edges and stairs, in their walking direction to reduce accidental falls. This system, known as electronic travel aid (ETA), is usually attached to a white cane and assists the user to avoid collisions and falling down. Utilizing recent advancements in information technology, we have introduced a smartphone as an ETA. The proposed fall risk alert system detects fall risk areas in the walking direction of the user based on a depth image obtained by a smartphone. The system calculates the fall risk based on the shortest distance to the edge of a platform and generates an alert as vibration from the smartphone or smartwatch. We conduct verification experiments using the smartphone-based fall risk alert system on a train platform in collaboration with visually impaired persons, who walk according to scenarios expected in daily life. The experimental results demonstrate that the system generates fall risk alerts on station platforms. Based on the walking paths and questionnaire answers, the proposed system is considered to provide similar information with tactile paving.

This study proposes a dual-purpose system for remote medical support during disasters and trauma treatment education, aiming to enhance procedural guidance and training effectiveness in resource-limited and high-risk environments. The system consists of a remote support system for trauma treatment and a medical education system for trauma training. The remote support system enables trauma experts at distant locations to interact with a 3D digital twin of disaster victims and their surroundings within a virtual reality environment, thereby providing procedural guidance to on-site medical personnel. The system uses widely available devices such as smartphones for 3D scanning and head-mounted displays for immersive visualization. This design enables rapid deployment at disaster sites without requiring specialized equipment or complex setup procedures. The trauma education system records expert surgical treatments using motion-capture technology and reconstructs them as interactive 3D avatars, allowing trainees to observe and learn techniques from multiple perspectives. The remote support system was evaluated through fingertip-based interaction tasks in a simulated disaster scenario, where the alignment accuracy was assessed using augmented reality overlays, resulting in a measured average error of 27.0 mm. Similarly, the trauma education system was evaluated for positional accuracy in instrument handling tasks. These results confirm the feasibility and practicality of the proposed system and demonstrate its potential to improve both emergency medical response and surgical education.

Accurate position estimation of unmanned aerial vehicles (UAVs) is essential for reliable and efficient task execution, such as the automated inspection of structures. However, localization is a significant challenge for UAVs in GNSS-denied spaces, such as under bridges, where satellite signals are obstructed. To address this issue, we employed a cooperative localization system consisting of a work-UAV, which performs tasks in the GNSS-denied space, and support-UAVs, which relatively localize the work-UAV. The critical challenge in this approach is the degradation of the GNSS-localization accuracy of the support-UAV owing to multipath effects, which are difficult to predict. This paper proposes an adaptive formation control method that enables support-UAVs to escape from multipath-affected spaces. Instead of focusing on signal processing or sensor fusion under multipath effects, our proposed system controls support-UAVs to escape from multipath-affected spaces where accurate GNSS-localization is difficult. This physical avoidance strategy enhances the robustness of the cooperative localization systems. Simulation results demonstrate the effectiveness of the proposed method. In two distinct scenarios, the average time spent by the support-UAV in multipath-affected spaces was reduced by 83.7% and 79.9%, respectively. Furthermore, the average task completion time was reduced by 37.6% and 14.3%, respectively.

The motion patterns typically exhibited by a robot manipulator during a pick-and-place task differ from those by a human. Humans often perform that movement in a manner that resembles a collision while still placing the object safely. This research proposes implementing a fast but rough contact motion by utilizing a series elastic actuator (SEA) on the robot’s joints to introduce compliance. Exploiting the impact mitigation of the SEA allows for high contact velocities while keeping safe values of contact force. The proposed method refrains from controlling the link’s actual position and utilizes contact with the environment to actively attenuate oscillations. The contact forces are limited and attenuated by designing the target position of the motor, joint stiffness, and velocity of the link. Simulation and preliminary results verify the capability of the SEA in mitigating impacts and limiting contact forces. In addition, the influence of each variable over the contact force and settling time is analyzed. Improvements over previous work on a variable-stiffness SEA mechanism based on compressed air are implemented. The proposed method is evaluated through simulation and experiments with a one-link arm. Results demonstrated the feasibility of the method, however, parameter settings need to be investigated.

The inherent nonlinear characteristics of friction adversely affects the control accuracy of joint motor drive systems in lower limb rehabilitation robots. Recognizing this challenge, this study proposes an improved friction model and further designs a feedforward compensation control scheme to mitigate motor friction on the basis of the friction model. Compensating for motor control utilizing the friction model, typically the Stribeck friction model, is a promising solution. To overcome the inherent limitation discontinuities of the Stribeck friction model, this study introduces an improved friction model by incorporating the sigmoid function into it. The friction parameter of the model is identified based on the data collected during the experiment, specifically the motor velocity and current. And to enhance the precision of the parameter identification, Kalman filtering algorithm is applied to mitigate the Gaussian noise generated during the experiment. Subsequently, the firefly algorithm is employed for offline identification and curve fitting of the friction parameters in the improved model. Based on the improved friction model, a feedforward compensation controller is further designed by integrating the traditional three-closed-loop PID motor control method with real-time friction compensation. The system employs humanoid gait patterns as input signals to achieve precise position tracking of the robot’s joint motors. Compared with the conventional PID control, the proposed feedforward compensation control reduces both position and current tracking errors. These results confirm that the feedforward compensation strategy, based on the refined Stribeck friction model, effectively mitigates the adverse effects of nonlinear friction, thereby improving the control performance of joint motor drive systems.

Here, we propose a deep learning-based two-stage recognition system for fruit-level crack classification in cherry tomatoes. This targets harvesting and sorting scenarios in real-world cultivation environments where leaves and stems are present. Cherry tomato cracking exhibits substantial visual variability, ranging from clearly split fruits to subtle white linear cracks around the calyx region. Therefore, crack-region-based or bounding-box-driven detection methods are highly susceptible to external noise, such as occlusions caused by leaves and stems, and illumination variations. This can strongly impair their generalization performance in field conditions. The wide diversity of crack appearances makes it difficult to collect sufficiently large and stable annotated datasets for robust training. To alleviate data scarcity, synthetic data generation was used to support model pre-training. Crack recognition in real-world environments was formulated as a two-stage framework comprising fruit detection followed by fruit-level crack classification. In the first stage, cherry tomatoes are detected using a You Only Look Once (YOLO)-based object detector. In the second stage, the detected fruit instances are classified as cracked or non-cracked through image-level classification using a Siamese network. Based on real-world environmental images, the proposed method achieved a crack classification accuracy of approximately 88% for red cherry tomatoes and successfully detected red cherry tomatoes, demonstrating its effectiveness for fruit-level crack differentiation under practical cultivation conditions.

In Japan, the quantity of domestically produced fruit has been gradually decreasing, while wholesale prices have continued to rise due to declining production volumes and a shift toward high-quality varieties. To address these trends, improving quality and reducing labor through automation have become urgent challenges. In precision viticulture, monitoring the growth of grape clusters plays a key role in yield estimation, disease management, and optimal harvest timing. Although recent advances in deep learning and 3D reconstruction have enabled accurate fruit detection and modeling in vineyards, tracking the same clusters on different days remains challenging because of branch movement, fruit growth, and varying imaging conditions. This study proposes a branch-based 3D alignment framework for the cross-day tracking of grape clusters. Stable vine structures, such as trunks and main branches, are reconstructed using Structure from Motion, and their spatial correspondences are estimated through SIFT-based matching and similarity transformation. Once the coordinate systems of different days are aligned, the grape clusters detected by CenterNet are associated based on spatial proximity in the unified 3D space. Experiments over multiple observation days demonstrated that the proposed method successfully maintained the consistent tracking of grape clusters throughout the growth period. These results indicate that branch-based alignment effectively stabilizes multi-day observations and facilitates the temporal monitoring of fruit growth, supporting automated phenotyping and future field robot applications in viticulture.