Ensuring the safety and efficiency of Autonomous Vehicles (AVs) necessitates highly accurate perception, especially for lane detection and lane-change manoeuvres. Among object detection frameworks, “You Only Look Once” (YOLO) algorithms have emerged as prominent contenders due to their rapid inference and commendable accuracy. However, the broad spectrum of YOLO variants and their applications in complex, real-world environments remain insufficiently mapped, necessitating a more integrative and critical perspective than what is typically offered by surveys. This comprehensive review synthesizes theoretical foundations, architectural innovations, and empirical evaluations of YOLO-based algorithms in AV-related tasks. It not only highlights key findings—such as the notable gains in real-time detection and adaptability to a range of driving conditions—but also explicitly identifies persistent gaps and limitations. These include difficulties in detecting subtle or degraded lane markings, handling unpredictable environmental factors like adverse weather and varied lighting, mitigating adversarial perturbations, and scaling effectively across diverse datasets and geographic regions. By critically examining these vulnerabilities, we illuminate the opportunities for refining YOLO's training paradigms, optimizing model architectures, incorporating sensor fusion, and fostering universally applicable datasets. The implications of addressing these gaps extend beyond mere technical refinements. Proactively tackling YOLO's current challenges can expedite the realization of safer, more robust, and globally adaptable AV navigation systems. In doing so, this review provides clear, actionable insights for researchers, engineers, and policymakers, guiding them toward strategic innovations that will strengthen AV perception and contribute to more reliable, future-ready transportation solutions.

Dropped Head Syndrome (DHS) presents significant challenges for patients in lifting or moving their heads. Physio-therapy can play a crucial role in maximizing patients’ neck function and mitigating symptom progression. This work focuses on the design, modeling, and control of a cable-driven robot for neck rehabilitation. The proposed mechanism is novel in how it employs cable-driven power transmission and multiple onlink joints for neck therapy to control and train the natural range of head-neck motions. Additionally, force-field control for an assist-as-needed robotic strategy is designed to prioritize human neck comfort and safety, as well as adaptable trajectory tracking for therapeutic neck exercises. Mathematical models, closed-chain kinematics, and Jacobian matrices are presented for both the robot-in-charge and the patient-in-charge. Feasible cable tensions are determined by solving a linear programming problem that accounts for the robot’s cable structure matrix and wrench-closure workspace. The robot prototype and control system are showcased, with initial experiments on healthy subjects confirming the device’s ability to cover the complete range of head-neck motions. Subsequent experiments with a humanoid neck demonstrate the robot’s performance in resistive and active modes, addressing jerk, force-moment requirements, and trajectory tracking.

This paper presents a principled framework for handling merged measurements in labeled random finite set (RFS) multi-object tracking, with a focus on directional audio cues from microphone arrays. In dense traffic environments, direction-of-arrival (DOA) estimation methods such as MUSIC often produce merged spectral peaks when multiple sound sources are in close angular proximity. This leads to ambiguous or missing measurements, undermining the performance of labeled RFS trackers like the labeled multi-Bernoulli (LMB) filter—particularly in terms of track continuity, cardinality estimation, and identity maintenance. To address this challenge, we propose a merge-aware tracking approach: (i) a survivor-merge compensation mechanism that detects missing targets due to DOA collapse and reconstructs their measurements using likelihood-based proximity; (ii) a birth-merge handling strategy based on likelihood-ratio testing and temporal evidence accumulation to recover new targets concealed within merged regions; and (iii) an LMB-consistent re-update step to integrate the generated synthetic DOAs without violating RFS principles. Following the audio re-update, the updated posterior is fused with a visual update—subject to occlusion and line-of-sight limitations via Generalized Covariance Intersection (GCI), enabling robust multi-sensor estimation under partial observability. We evaluate the proposed method in traffic scenarios with close-proximity vehicle interactions and angular congestion, simulated using realistic urban layouts and dynamic traffic conditions. Monte Carlo experiments demonstrate significant improvements in trajectory continuity, cardinality accuracy, and OSPA metrics over PHD-GCI and LMB-GCI baselines. While demonstrated on DOA-based sensing, the method generalizes to other modalities susceptible to measurement merging, such as LiDAR, radar, and vision.

This paper proposes a novel multi-object tracking framework for intelligent vehicles, addressing the challenge of maintaining situational awareness during visual occlusions, where objects may be fully or partially blocked by other vehicles, infrastructure, or environmental elements. We first evaluate a visual-only tracking pipeline using static birth modeling and extend it with adaptive birth to compensate for missed detections; however, both approaches struggle to maintain tracks through occlusion events. Extending our previous work on audio-visual fusion, this study introduces a non-traditional, data-driven audio likelihood function that operates directly on raw microphone array signals in a track-before-detect (TBD) manner. Unlike conventional approaches relying on preprocessed direction-of-arrival (DOA) estimates, the proposed likelihood is computed using the Multiple Signal Classification (MUSIC)-based spatial spectrum analysis and formulated as a Gaussian mixture. Audio measurements are incorporated into a Sequential Monte Carlo (SMC) implementation of the Labeled Multi-Bernoulli (LMB) filter, and the resulting visual and audio posteriors, obtained after independent LMB updates, are fused using Generalized Covariance Intersection (GCI). A physics-based simulation framework is developed to generate realistic vehicle acoustic signals and evaluate performance under two urban traffic scenarios with varying interaction density (moderate and dense). Monte Carlo experiments demonstrate that the proposed approach consistently improves tracking accuracy, cardinality estimation, and robustness compared to visual-only baselines. The results highlight the effectiveness of signal-level audio-visual fusion in maintaining track continuity under prolonged occlusions and increased scene complexity.

Deterministic reactive control (DRC)-based obstacle avoidance has become a foundational capability for autonomous robotic systems operating in modern manufacturing environments, where robots must interact safely with humans, machines, and dynamic surroundings while maintaining high productivity. DRC approaches compute control actions directly from real-time sensory feedback, enabling fast, predictable, and computationally efficient responses. Despite extensive research, DRC methods have largely been treated as isolated techniques, and a unified, up-to-date review of their theoretical foundations, practical behavior, and industrial relevance has been lacking. This paper presents a comprehensive review of DRC-based obstacle avoidance strategies, with a specific focus on collaborative robots (cobots) and autonomous mobile robots (AMRs) in manufacturing applications. We develop a systematic taxonomy based on underlying control principles, including rule-based reactive control, potential-based gradient methods, velocity-based reactive methods, constraint-modulated dynamical systems, and primitive-embedded reactive systems. The review analyzes these approaches under different obstacle scenarios, including human-like obstacles, known environments, and unknown or partially observable environments, and evaluates their suitability for safety-critical industrial deployment. In addition, the paper reviews real-world deployments and industrial case studies that highlight the strengths of DRC methods in real-time responsiveness, determinism, and safety reasoning. Key challenges limiting large-scale adoption, such as safety-throughput trade-offs, deterministic obstacle modeling, crowd navigation in factories, Sim-to-Real transfer, and maintainability after mass deployments, are critically discussed. Finally, future research directions are identified, including integrating DRC with control barrier functions, safe learning frameworks, and modular design automation frameworks, to enable the next generation of safe and reliable autonomous robotic systems for smart manufacturing.

The autonomous search for an unknown radio frequency (RF) emitter in an unmapped indoor environment is a fundamental capability of intelligent robotic systems, enabling applications such as search and rescue, surveillance, environmental monitoring, biological studies, and industrial automation. To address this challenge, this article proposes a theoretical framework for autonomous RF emitter localization and search in structured yet previously unmapped indoor environments. The searching agent fuses the nondirectional received signal strength (RSS) measurements to estimate the RF emitter position and signal strength, while concurrently mapping the environment and localizing itself using LiDAR measurements. The task of simultaneous emitter localization, mapping, and agent self-localization is formulated within the sequential Bayesian framework. The main technical challenge lies in the unknown propagation path losses affecting the RSS measurements, which is addressed using an imprecise likelihood model based on the interval valued path loss exponent (PLE). The sequential Bayesian estimation is implemented using a Rao-Blackwellized particle filter (RBPF) for simultaneous localization and mapping (SLAM) and, conditional on its result, a separate Monte Carlo estimator of the RF emitter parameter vector. Path planning for search motion control is based on an entropy-reduction reward in the RF emitter parameter space; its performance is assessed in simulation against two heuristic search strategies. A physical robotic platform was developed to experimentally validate the theoretical framework. Extensive simulations and real-world experiments demonstrate the effectiveness of the proposed search algorithm.

Unmanned Aerial Vehicles (UAVs) integrated with lightweight visual cameras hold significant promise in renewable energy asset inspection and monitoring. This study presents an AI-assisted soiling detection methodology for inspecting solar photovoltaic (PV) panels, using UAV-captured RGB images. The proposed scheme introduces an autonomous end-to-end soiling detection model for common types of soiling in solar panel installations, including bird droppings and dust. Detecting soiling, particularly bird droppings, is critical due to their pronounced negative impact on power generation, primarily through hotspot formation and their resistance to natural cleaning processes such as rain. A dataset containing aerial RGB images of PV panels with dust and bird droppings is collected as a prerequisite. This study addresses the unique challenges posed by the small size and indistinct features of bird droppings in aerial imagery in contrast to relatively large-sized dust regions. To overcome these challenges, we developed a custom model, named SDS-YOLO (Soiling Detection System YOLO), which features a Convolutional Block Attention Module (CBAM) and two dedicated detection heads optimized for dust and bird droppings. The SDS-YOLO model significantly improves detection accuracy for bird droppings while maintaining robust performance for the dust class, compared with YOLOv5, YOLOv8, and YOLOv11. With the integration of CBAM, we achieved a substantial 40.2% increase in mean Average Precision (mAP50) and a 26.6% improvement in F1 score for bird droppings. Dust detection metrics also benefited from this attention-based refinement. These results underscore the CBAM’s role in improving feature extraction and reducing false positives, particularly for challenging soiling types. Additionally, the SDS-YOLO parameter count is reduced by 24%, thus enhancing its suitability for edge computing applications.

A 2D laser profiler is commonly utilized in high-precision robotic settings to capture detailed surface profiles for 3D scanning. By collecting and combining numerous such measurements from different viewpoints, it is possible to assemble a comprehensive 3D map. However, to effectively merge these individual 2D profiles into a singular global framework, the spatial relationship between the scanners and the robot's reference frame is required. Traditional hand-eye calibration techniques typically necessitate specific calibration artifacts or extraneous positional sensors, and the process is either manually executed or only partially automated, demanding considerable time and effort. This paper introduces an innovative, closed-form approach to hand- eye calibration that can be applied to a single scanner or an array of multiple scanners. Our method circumvents the requirements for initial parameter estimates or specialized calibration implements, instead employing a flat plane for hand-eye calibration. This method paves the way fora fully automated calibration sequence comprising only three rotational and three translational poses, reducing the total calibration duration. This streamlined process has undergone strict experimental validation utilizing a calibrated sphere, proving its effectiveness not only with a solitary scanner setup but also with an ensemble of three scanners.

This paper addresses a gap between the capabilities and utilisation of robotics and automation in laboratory settings and builds upon the concept of Self Driving Labs (SDL). We introduce an innovative approach to the temporal characterisation of materials. The article discusses the challenges posed by manual methods involving established laboratory equipment and presents an automated hyperspectral characterisation station. This station integrates robot-aided hyperspectral imaging (HSI), complex material characterisation modelling, and automated data analysis, offering a non-destructive and comprehensive approach. This work explains how the proposed assembly can automatically measure the half-life of biodegradable polymers with higher throughput and accuracy than manual methods. The investigation explores the effect of pH, number of average molecular weight (Mn), end groups, and blends on the degradation rate of polylactic acid (PLA). The novel contributions of the paper lie in introducing an adaptable classification station for characterisation and presenting an innovative methodology for polymer degradation rate measurements. The proposed system holds promise for expediting the development of high-throughput screening and characterisation methods within advanced material and chemistry laboratories.

Additive manufacturing (AM) is an emerging digital manufacturing technology to produce complex and freeform objects through a layer-wise deposition. High deposition rate robotic AM (HDRRAM) processes, such as cold spray additive manufacturing (CSAM), offer significantly increased build speeds by delivering large volumes of material per unit time. However, maintaining shape accuracy remains a critical challenge, particularly due to process instabilities in current open-loop systems. Detecting these deviations as they occur is essential to prevent error propagation, ensure part quality, and minimize post-processing requirements. This study presents a real-time monitoring system to acquire and reconstruct the growing part and directly compares it with a near-net reference model to detect the shape deviation during the manufacturing process. The early identification of shape inconsistencies, followed by segmenting and tracking each deviation region, paves the way for timely intervention and compensation to achieve consistent part quality.

This paper presents an innovative approach to improving situational awareness in autonomous vehicles by leveraging audio-visual fusion for enhanced object detection, particularly in urban environments prone to visual occlusions. Traditional autonomous vehicles situational awareness systems face limitations in complex, crowded settings where occlusions by other objects or environmental factors obstruct visual sensors. We propose a fusion of audio cues with visual data to overcome occlusions, utilizing a Labeled Random Finite Set model integrated with audio features such as engine sounds and horn signals. Our approach demonstrates that combining visual and auditory information through multi-object tracking enables autonomous vehicles to maintain accurate awareness of their surroundings, even during prolonged occlusions. Results from simulated urban traffic scenarios validate the effectiveness of this fusion method, showing a significant reduction in tracking errors and enhanced resilience against occlusions.

Robotic additive manufacturing using a cold spray deposition head attached to a robotic arm can deposit material in a solid state with deposition rates in kilogrammes per hour. Under such a high deposition rate, the complicated interplay between the robot’s motion, gun standoff distance, spray angle, overlapping, and the interaction of supersonic powder particles with a growing structure could cause overabundance or deficiency of material build-up. Over time, the accumulation of these discrepancies can negatively affect the overall shape and size of the final manufactured object. In-process spatio-temporal 3D reconstruction, also known as 4D reconstruction, could allow for early detection of deviations from the design, thus providing the opportunity to rectify at an early stage, making the process more robust, efficient and productive. However, in-process model reconstruction is challenging due to the dynamic nature of the scene (e.g. sensor and object relative movements), the three-dimensional growth of a time-varying build object, the textureless nature of build surfaces, and its computational complexity. We propose a real-time, in-process 4D reconstruction framework for free-form additive manufacturing processes, such as cold spray that deals with a real-time dynamic and evolving scene built by incremental deposition of materials. In our approach, temporal point clouds from three cameras are acquired and segmented to extract the region of interest (build object). The subsequent multi-temporal and multi-camera registration of the segmented 3D data is addressed by combining geometrically constrained Fiducial marker tracking and plane-based registration without drift accumulation. Finally, the registered point clouds are fused via voxel fusion of growing parts to reconstruct the 3D model of the object with smoothened surfaces. The proposed solution is deployed and verified in a robotic cold spray cell with different test scenarios and shape complexities.

We propose a hybrid adaptive control method with robust disturbance rejection for a novel personal micromobility vehicle that uses ballbot drive mechanics with a single spherical wheel. Intelligent control of such a micromobility vehicle is significantly sensitive to discrete events of short duration and high-intensity road disturbances and shocks. Sudden bumps, potholes, pavement or kicks typically cause these events. Our approach combines adaptive disturbance observer (DOB) with two other techniques to achieve better and more robust nonlinear control of a ballbot drive vehicle (BDV) against small to large external disturbances. The proposed hybrid DOB-Based combined control system (DOB-CCS) stabilizes the body position via a DOBbased partial feedback linearization (PFL) strategy while enabling ball movement on the floor by introducing a DOB-based sliding mode control (SMC). The control scheme consists of linearly combined components responsible for maintaining an upright posture about unstable equilibrium points and tracking the motion under disturbances. Simulations, real-platform experiments and respective comparative studies show the efficacy of the proposed method for a BDV to reject external disturbances. The practical performance of the proposed DOB-CCS strategy is experimentally validated and compared with a CCS method. The experimental results indicate less than 0.7 and 1.5 degrees of ballbot's orientation error and robust rejection of disturbances under kicks during balancing and transferring modes, respectively

Ballbots are omnidirectional self-balancing platforms that can be exploited in many applications to detect, track, or interact with objects or humans, such as a service robot. Ballbot will enable mobile robots to stand tall and move elegantly through busy environments. However, maintaining equilibrium through synchronization of motion between the ball and the body of a Ballbot is still an open research problem. This article presents a synchronization control (SC) design, with synchronization and coupling errors for Ballbots to stabilize the body and control ball transfer simultaneously. The proposed SC method is applied to the two 2-D planar models of a Ballbot robot. The dynamic model of the Ballbot is derived, and parameters are identified online using the intelligent particle swarm optimization method. The proposed controller is proven to guarantee asymptotic convergence to zero errors in tracking and synchronization. The stabilizing and transferring problems are investigated through several simulations and experiments by using an actual Ballbot platform. Moreover, the controller performance is compared with an augmented proportional derivative controller and a partial feedback linearization controller. The results and comparisons demonstrate a superior stabilization accuracy of the proposed SC method.

A frequent and accurate quality inspection procedure to assess the quality requirements during the life cycle of buildings is crucial. Among different quality measures, the dimensional quality that involves spatial features of buildings is of significant importance. However, the traditional manual inspection of dimensional quality in buildings is unreliable and tedious. Thus, this study presents an end-to-end method for quality inspection of building structural members using point cloud datasets. The proposed method, first, detects and labels structural members within the point cloud based on a set of domain-specific geometric and semantic definitions. Then, each structural member's section width, height, and length are obtained with the proposed bounding box method. Experiments on three real-world buildings' point clouds with various geometric features and noise levels, occlusion, and outliers were also conducted, illustrating the performance efficiency and accuracy of the proposed model for dimensional quality inspection of building structural members.

The growing environmental concerns have emerged the necessity of sustainable waste management of construction and demolition (C&D) wastes. This review explores the advancements in artificial intelligence (AI) and robotics to automate C&D waste sorting. A comprehensive examination of this domain is conducted by structuring the paper around six research questions. Current trends and potential future directions are revealed by performing methodology and data analysis involving bibliometric and scientometric studies. Notably, recent research emphasises circular economy, AI, and robotics, underscoring the importance to enhance AI for precise categorisation. The scarcity of publicly available datasets is a central challenge in the C&D waste domain, that hinders effective AI applications. However, data augmentation, data synthesis, generative AI, and transfer learning have been identified as crucial techniques to enhance dataset quality and categorization accuracy. While AI draws significant attention in the C&D waste domain, this review shows a lack of AI-enabled robotics systems due to the complex nature of waste sorting and collection. In summary, this study's findings highlight the need for new methods and techniques integrating multisensory fusion, unsupervised machine learning and robotics intelligence to continuously learn and adapt to new waste streams and materials, making them highly efficient in sustainable waste management.

Building structural works require regular inspections and monitoring of damage progression to ensure compliance with relevant standards and jurisdictional requirements. Conventional quality inspections rely on manual measurements, which is costly, tedious, and error-prone. Recently, Terrestrial Laser Scanners (TLSs) have shown promising performance in terms of accuracy, cost, and efficiency for inspecting building structural members. Nevertheless, utilizing TLS for quality inspection of building structural members lacks generalizable and efficient approaches to inspect and monitor various quality criteria for different types of building structural members over time. To fill this gap, a generalizable framework for inspecting and monitoring building structural members using multi-temporal point clouds is developed in this study. First, an informative cross-section shape and structural member type invariant representative plane from each building structural member, which preserves the underlying dimensional imperfection-related features, is extracted. Then, a combination of geometric imperfections in building structural members, including deflection and slope in beams and inclination and straightness in columns adopting the standard practice and definitions in building codes and standards, are identified and quantified. Finally, a change detection method is proposed to monitor the geometric quality of building structural members over time. Experiments on real-world multi-temporal point clouds of a building under renovation are performed to validate the performance of the proposed framework by comparing the calculated deformations with the field measurements. An average Mean Ab-solute Error (MAE) of 1.59 mm & PLUSMN; 0.72 mm and an average MAE of 0.67 mm & PLUSMN; 0.25 mm were reported for building structural member deflection and slope deviation, respectively, compared to manual measurements. The basis of the presented methodology framework, including represen-tative plane detection and compliance checks criteria, can be extended for various geometric quality compliance checks for different building structural members.