The advancement of discrete manufacturing systems has led to increasingly intricate interactions among machines, workers, and fixtures. In modern workshops, these three resource types are not independently assigned but interact with complex coupling logic during manufacturing process. However, most existing studies consider either machine–worker or machine–fixture collaboration and do not address the joint coordination of machines, workers, and fixtures, posing challenges for efficient decision-making in actual manufacturing environments. To address this issue, this paper proposes a flexible job-shop scheduling problem with integrated machine–worker–fixture collaboration constraints (FJSP-MWF). A novel two-tuple encoding method is proposed to explicitly model job-fixture and machine–worker interactions, and a resource-limited decoding method is developed to ensure feasible schedules. Furthermore, a coevolutionary optimization algorithm (BEA-RLS) is proposed by combining a bi-population genetic algorithm with four resource-based local search strategies. Experiments on real-world-inspired instances demonstrate that the proposed method achieves superior performance, particularly on medium-scale instances, with over 20% improvements in total weighted tardiness (TWT) compared to existing benchmarks. A real-world case study illustrates that the TWT and makespan of BEA-RLS are reduced by 25.86% and 35.48%, compared to the expert experience.
Incorporating multiple robotic manipulators into large-scale manufacturing systems enhances production efficiency and expands manufacturing capabilities beyond those of single-robot systems. Workpiece positioners in robotic manufacturing have demonstrated significant benefits for process optimization, but coordination strategies for multi-robot systems with shared positioners have received limited attention. This work presents a task-space coordinated trajectory-tracking control framework for multi-robot manufacturing systems, in which robots coordinate their motions within a shared, dynamic workpiece positioning frame. A workpiece positioner actively adjusts the pose of the manufactured component to enable greater operational concurrency and improve overall production efficiency. The proposed motion-coordination scheme employs a distributed and scalable architecture, supporting coordination across heterogeneous multi-robot systems. Two optimization methodologies are introduced to manage kinematic redundancies and maintain continuous, near-optimal operation throughout the manufacturing process. The first strategy exploits a task-space dimensionality reduction to achieve locally optimal configurations by leveraging symmetry-axis rotations of the tool. The second strategy utilizes the workpiece positioner to drive the coordinated robots toward stable and kinematically favorable configurations. For both optimization strategies, multiple objectives are defined to improve key performance metrics, including manipulability, configuration consistency, proximity to mechanical limits, and motion efficiency. Addressing a key limitation of existing coordination approaches, the framework is designed around online setpoint modification, allowing coordinated robots to respond effectively to in-situ process feedback. The proposed control framework is validated using the Robot Operating System (ROS) middleware on a combination of physical and simulated multi-robot system hardware.
The increasing demand for atomic and close-to-atomic scale manufacturing of advanced optical components poses significant challenges for polishing hard and brittle materials such as silicon carbide (SiC) ceramics. Photocatalysis-assisted polishing has demonstrated high efficiency and surface quality in planar processing, while its application to complex surfaces remains limited. To address this issue, a five-axis photocatalysis-assisted polishing system is developed to enable the machining of complex geometries. Two motion generation strategies are proposed based on this developed platform, namely a dimension-reduced motion generation approach for rotational-symmetrical surfaces and a homogeneous-transformation-based motion generation approach for freeform surfaces. Polishing experiments on complex surfaces are conducted and the results demonstrate stable material removal and satisfactory surface quality, with a surface roughness of 0.22 nm in Sa. The study shows that the proposed motion generation strategies, combined with photocatalysis-assisted polishing, provide an effective approach for high-precision finishing of complex SiC ceramic surfaces.
An effective stiffness evaluation method and its associated indices are crucial for ensuring high-precision performance of parallel kinematic robots (PKR) under heavy and time-varying loads. Traditional stiffness indices, derived from algebraic properties of the stiffness matrix, such as eigenvalues, condition number, and trace, quantify global or directional stiffness but fail to capture the geometric structure and deformation variation patterns under complex loading. To address this, we propose a stiffness evaluation method grounded in the intrinsic geometry of Riemannian manifolds. By mapping the PKR’s deformation space onto a coordinate-invariant manifold, the method intrinsically accounts for directional coupling and deformation anisotropy. An overall compliance matrix is first constructed using conservative congruence transformation, integrating deformations of joints, links, and the end-effector. A normalized force space is then introduced, modeling deformations as a hyper-ellipsoid embedded in high-dimensional space. Within this framework, scale curvature and its variance are employed as principal indices to jointly characterize stiffness magnitude and robustness. A 6-PSS PKR is analyzed under three representative loading scenarios using the NSGA-II multi-objective optimization method. The curvature-based optimization reduces average maximum deformation errors by over 30 % compared to traditional single-load optimization. A new prototype 6-PSS PKR forming machine with a 6 MN payload has been developed, and forming experiments under distinct conditions demonstrate a maximum deviation of only 11 % from theoretical predictions. The optimized configuration reduces deformation errors by over 35 %. These results validate the effectiveness of the proposed evaluation method, significantly enhancing the PKR’s ability to withstand complex and varying external loads.
In manufacturing engineering, process planning transforms engineering drawings into manufacturing process plans by determining appropriate machining operations and their sequences, thereby directly influencing production efficiency and product quality. Traditional process planning relies heavily on expert experience, resulting in low efficiency and limited knowledge standardization. Although computer-aided process planning (CAPP) methods have been developed, they typically depend on handcrafted rules or structured inputs, limiting their ability to handle complex semantics, diverse drawing styles, and cross-task generalization. Vision-language models (VLMs) provide a promising paradigm for drawing understanding. However, general-purpose VLMs often lack specialized manufacturing knowledge and struggle to align high-level representations with pixel-level details in high-resolution drawings. This leads to the loss of critical information and degrades the quality of generated process text. We propose ProcessMM, a domain-specialized multimodal model for manufacturing process planning. ProcessMM adopts hierarchical vision-text alignment with dual-channel features. By incorporating OCR text into visual representations via an OCR-guided attention (OGA) module, ProcessMM enhances fine-grained detail perception. We also develop a sentence-level attention attribution method that provides process planners with visual evidence for generated process text. Evaluations on three engineering datasets collected from real-world manufacturing processes show that ProcessMM consistently outperforms dominant multimodal large language models (MLLMs) in manufacturing process-text generation, demonstrating effective and efficient automated process planning from engineering drawings.
Wire harness assembly exemplifies a non-rigid object assembly task that remains challenging to automate due to object deformability, occlusions, high variability, and workspace constraints. This study introduces a vision-based human–robot collaboration (HRC) framework developed for wire harness assembly in automotive final assembly processes. In this system, the robot performs repetitive, force-intensive tasks, while the human operator handles operations that require dexterity and adaptive decision-making. The proposed approach combines marker-based pose estimation of wire harness components with a hand-triggered control scheme. Detected hand landmarks define a region of interest, enabling intentional, context-aware robot activation. The HRC framework is validated by performing wire harness installation on the vehicle chassis. A two-phase within-subjects experimental study compares manual installation with HRC-assisted installation in both laboratory and industrially relevant environments, corresponding to Technology Readiness Level 4 to 6. The results indicate that robotic assistance significantly reduces localized physical discomfort and physical demand while maintaining a high success rate in wire harness installation. However, overall workload does not differ significantly between HRC and manual conditions. In the HRC condition, both mental demand and average assembly time increase significantly compared to manual assembly. Cycle-time analysis reveals that the robot execution phase accounts for the largest proportion of total time in the collaborative workflow. These findings suggest that vision-based HRC can provide targeted ergonomic benefits for non-rigid object assembly, while also introducing cognitive and temporal trade-offs. Therefore, real-world deployment requires further improvement in interaction fluency and throughput. The proposed method and code are available at https://github.com/HWANG7308/ClampTracking.
Embodied intelligence provides a promising paradigm for robotic machining systems, enabling autonomous perception, reasoning, and execution of diverse tasks via tight coupling of physical embodiment and decision-making. However, its industrial applications remain limited, as machining performance is highly constrained by robot physical properties and process conditions. Accordingly, this paper innovatively proposes an embodied intelligence control framework for hybrid robot machining systems. The framework decomposes complex machining tasks into three phases, including positioning, planning and manipulating. Driven by meta-skills, the framework realizes autonomous decision-making and execution. Considering the hybrid robot’s structural features, a realistic digital twin simulation environment is originally built based on a model-data-physics framework, realizing real-time simulation of key physical attributes to fully characterize the system’s physical state and support meta-skill learning. Moreover, a reinforcement learning method integrated with the strong constraints of the mechanistic models is proposed. By introducing reward functions and constraint conditions constructed based on the mechanistic models to guide the learning process, the efficiency and stability of learning are improved. Two representative meta-skills are developed through the method: workpiece clamping position selection and collision-free smooth trajectory planning. Finally, an embodied intelligent machining system is established and validated through milling tests on three workpieces with distinct geometric features, as well as a drilling task to assess adaptability to hardware configuration changes. The results demonstrate that the framework autonomously completes the full decision-making and execution without retraining, indicating the application potential of the proposed meta-skill-driven embodied intelligence framework for flexible manufacturing.
Unexpected collisions during CNC machining execution remain a major cause of equipment damage. Offline CAM verification cannot detect collisions that happen during real machining. Sensor-based methods are costly, unreliable in harsh environments, and in practice are limited to triggering an emergency stop after contact. Reliable collision avoidance before contact without additional external sensors remains an open challenge. To bridge this gap, this paper proposes a simulation-driven active safety framework integrating predictive collision detection with autonomous retraction. First, a synchronized CNC digital replica of the physical machine is maintained via bidirectional communication; on this basis, a Predictive Shadow Model (PSM) projects tool motion ahead of physical execution to detect potential collisions within a look-ahead horizon. Second, upon detecting a risk, an efficiency-oriented hierarchical planner generates collision-free retraction paths. Third, the planned retraction path is converted into CNC-executable G-code, while an Online Trajectory Generation (OTG) algorithm propagates the corresponding kinematically constrained virtual trajectory for continuous predictive monitoring during retraction. Validation on a synchronized three-axis CNC testbed demonstrates an average communication latency of 1.16 ms and an aggregate trigger-to-hold response time of 136 ms, below the configured 0.5 s look-ahead horizon for predictive intervention. For typical local hazards, the Stage-I heuristic planner achieves an average retraction-planning latency of 28 ms, reducing latency by 95.4% relative to a Stage-II-only voxel-grid A∗ baseline. These results support the scenario-level feasibility of integrating simulation-driven predictive models with real-time CNC control for pre-contact active safety.
Digitalization initiatives have led to the widespread adoption of digital twins (DTs), particularly in manufacturing. The explosive growth of DTs makes us think about what the most abstract parts in building the DT systems are and how to promote the reconfigurability of DTs so that they can be agile to adopt the flexibility needs from Industry 4.0 manufacturing systems. However, the heterogeneous granularity and dynamic evolution of manufacturing systems present significant challenges to the reconfigurability of their corresponding DTs. To address these challenges, this paper presents MetaTwin, a reconfigurable DT framework as an integrated solution to abstract DTs into general domains. To be more specific, a novel Domain-Meta structure is first proposed, which addresses the granularity misalignment problem by decoupling the system into stable Domains and reconfigurable Meta units. Then, a dual-dimensional reconfiguration mechanism decomposes complex reconfigurations into macro and micro dimensions, achieved by composing Meta units and reconfiguring their internal pattern-driven Meta Models. A prototype-based case study involving the progressive reconfiguration from an assembly-only DT to a printing-and-assembly DT illustrates how MetaTwin supports reconfiguration scope identification, macro-level structural changes, micro-level behavioral adaptations, and runtime coordination. Furthermore, implementation-level indicators further suggest that the MetaTwin-based implementation localizes changes more effectively than a conventional implementation in the case setting.
Industrial robots are widely deployed in continuous trajectory tasks. However, inadequate accuracy caused by factors such as payload variations often necessitates lengthy commissioning cycles. Existing error compensation methods face practical limitations in continuous-trajectory applications, including switching between different inverse-kinematic branches, difficulty in preserving process-required orientation, and deployment mismatch with proprietary industrial controllers. To address these challenges, this paper proposes an error compensation method based on a Model-Informed Residual Network (MI-ResNet) and an axis-locking mechanism. The proposed method integrates a controller-consistent kinematic model with a residual network to achieve accurate error modeling under limited training samples. For inverse compensation, a model-agnostic black-box numerical inverse solver is constructed, in which an axis-locking strategy is introduced to suppress configuration switching while preserving process-compatible orientation. Experiments on multiple KUKA robots demonstrate that the mean positioning error is reduced to below 0.3 mm. In real-world automotive hemming tests, the mean 3D positioning error at the sampled trajectory points decreased by approximately 77%, with the maximum 3D positioning error consistently constrained within 0.5 mm, demonstrating improved positioning accuracy at sampled trajectory points and process-oriented orientation stability.
This paper presents an adaptive control strategy that enables the active use of a pneumatic spindle with passive radial compliance for robotic edge finishing of planar surfaces. The aim is to overcome cutting interruptions and unstable tool-workpiece interaction caused by trajectory-positioning errors, while exploiting the fast local response of passive compliance without requiring additional force sensors or extra actuation hardware. Unlike conventional rigid active-compensation architectures, the proposed approach combines passive compliance with online trajectory correction driven by process data. The method acquires robot motion data from the internal variables of the robot controller through its OPC UA server and combines them with the magnitude and direction of the radial compliance angle, together with spindle rotational speed measured by the pneumatic spindle. These signals are used to estimate workpiece contact in real time and compute a hybrid corrective action based on an incremental PI branch and a conditional PD boost. The control law is contact-dominant, while spindle speed acts as an auxiliary indicator of process load to prevent excessive tool deceleration and cutting interruption. Experimental validation on contouring trajectories with initial positioning deviations of up to ±2 mm showed that the proposed strategy maintained continuous contact within a stable spindle-speed regime, completely removed the initial small burrs, and produced a continuous, uniform chamfer smaller than 0.2 mm. These results demonstrate that passive pneumatic tooling can be used actively and effectively for robotic edge finishing despite positioning errors inherent to robotic machining.
Assembly of alignment-sensitive optical systems remains challenging because even small positional or angular errors can measurably affect system performance. In solid-state lasers, such errors can lead to changes in output power, beam quality, and stability. As a result, assembly of these systems remains largely manual, which increases production effort and cost, limits scalability, and often results in bulky designs constrained by large kinematic optic mounts. To address this challenge, this paper presents a feedback-guided robotic assembly workflow for compact, alignment-sensitive optical systems and validates it experimentally with a diode-pumped ruby laser. The workflow combines vision-guided pickup and passive placement with fluorescence-based axial positioning of the crystal, optical-feedback-based cavity alignment, stepwise optimization of the laser output, and permanent fixation via UV-curable adhesive. The process was implemented on a commercial precision optics assembly platform and executed without manual adjustment. As a representative validation case, the assembled ruby laser produced diffraction-limited output at 694.3 nm with an output power of 30 mW. The results show that passive placement, in situ optical feedback, optimization, and bonding can be integrated into a repeatable robotic process for laser assembly. The demonstrated workflow provides a practical route toward automated assembly of optical systems whose performance depends critically on precise alignment and on preserving that alignment during and after bonding.
Hand–eye calibration plays a critical role in robotics and computer vision, enabling accurate coordination between cameras and manipulators for perception-driven tasks. Ensuring high calibration accuracy is essential for reliable robotic performance across industrial, medical, and autonomous systems. This review aims to examine, classify, and assess state-of-the-art performance metrics in hand–eye calibration methods. A systematic literature review was conducted using a structured search strategy across major databases, guided by predefined inclusion and exclusion criteria and following the PRISMA framework. Titles, abstracts, and full texts were screened to ensure the selection of primary experimental studies, resulting in seventy relevant papers that collectively represent current methodological developments. The findings reveal that classical analytical approaches remain dominant due to their computational efficiency, while optimisation-based and hybrid frameworks consistently provide enhanced accuracy through iterative refinement. Emerging machine learning and deep learning methods exhibit promising results but lack robust benchmarking and generalisability. Significant variability was observed across eye-to-hand, eye-in-hand, and combined configurations, particularly in Position, re-projection, and translation & rotation error reporting, demonstrating a lack of standardised evaluation practices. Analysing performance metrics proved challenging initially, as studies presented calibration accuracy using diverse and inconsistent error types. This inconsistency underscores a broader gap in the field related to reproducibility and common reporting standards. This review synthesises methodological trends and provides a consolidated comparative analysis of seventy primary studies in accordance with their performance presentation. It highlights the need for unified benchmarking frameworks and proposes future research directions to advance accuracy and robustness in real-world calibration.
