To address the core drawbacks of traditional robot path planning algorithms in complex constrained environments, including local optima trapping, imbalance between global exploration and local exploitation, and difficulty in coordinating path feasibility with overall performance, this paper proposed a path planning method based on multi-strategy fused Non-Uniform Mutation Improved Grey Wolf Optimization (NUMIGWO). The algorithm integrated four core strategies: piecewise chaotic mapping for population initialization to improve solution space traversal uniformity; a nonlinear decreasing dynamic weight mechanism to adaptively switch between global exploration in early iterations and fine local exploitation in later stages; an adaptive Cauchy- Gaussian hybrid mutation with four parameters (r, α, β, γ) to enable both large-span local optima escaping and fine-grained search; and a path feasibility detection and repair strategy to explicitly embed environmental constraints for reliable obstacle avoidance. Verified by six standard test functions, NUMIGWO significantly outperformed Traditional Grey Wolf Optimization (TGWO) and Particle Swarm Optimization(PSO) in terms of optimal value, mean value, standard deviation and local optima escaping capability. In the application of an automotive engine block workshop, the optimal parameter combination was obtained through L25(5 4 ) orthogonal experiments. The proposed method outperformed TGWO, A*, and RRT in comprehensive performance. Compared with the above algorithms, NUMIGWO reduced the path length by 6.02%, 1.34%, and 20.55% and decreased the average turning angle by 7.29%, 43.56%, and 64.42%, respectively. It achieved a minimum safety distance of 5.64 m and a larger average obstacle distance than A* and RRT, which effectively improved the overall practicability and reliability of robot path planning for intelligent manufacturing applications.
There are industry pain points in the visual design of humanoid robots, such as creative reproduction deviation, distorted appearance, and incompatibility between aesthetic design and engineering structure. This paper proposes a visual collaborative creation method for CAM DiT humanoid robots based on generative artificial intelligence, constructing a dual driven overall architecture of visual creativity generation and engineering constraint control. The method relies on a multimodal visual information encoding alignment mechanism, integrates text style semantics and sketch geometric features, and accurately restores design creativity. By embedding 12 core engineering constraints into the DiT generation model, the integration of appearance aesthetics and motion feasibility can be achieved. At the same time, establish a feasibility verification feedback mechanism and a human-machine collaborative iterative strategy for preference scoring and local mask correction, to achieve precise optimization of visual design. This paper constructs an exclusive dataset for robot vision design, adding dual evaluation indicators of visual aesthetics and engineering constraints. The effectiveness of the method is verified through comparative experiments, module ablation experiments, sensitivity analysis, and user experience testing. The results indicate that this method can effectively improve the quality of robot vision generation and collaborative creation efficiency, providing new technical support for the intelligent and engineering implementation of humanoid robot vision design.
Humanoid robots increasingly rely on multimodal perception and closed-loop interaction to perform contact-rich industrial assembly. However, existing vision-language-action methods often lack fine-grained contact-semantic understanding, explicit structural consistency modeling, and reliable recovery from disturbances. This paper proposes EFSC-Learn, an embodied feedback-driven framework that integrates RGB-D vision, tactile sensing, force-torque signals, proprioception, and human feedback. The method combines temporal alignment, reliability-aware multimodal fusion, contact-semantic extraction, graph-based structural consistency learning, recurrent structural memory, and safety-constrained action projection to generate corrective actions. Experiments on four public robotic manipulation datasets compare EFSC-Learn with ACT, Diffusion Policy, Octo, and OpenVLA. The results show that the proposed method consistently improves contact-semantic recognition, structural relation prediction, assembly success, perturbation recovery, correction accuracy, and contact safety while retaining practical online efficiency under heterogeneous sensing conditions and contact uncertainties. These findings demonstrate that explicit multimodal structure reasoning provides a robust basis for closed-loop humanoid robot manipulation in complex industrial environments.
Industrial humanoid robots require robust visual-tactile embodied intelligence to perform precision assembly under occlusion, contact uncertainty, and human-robot coexistence. Existing visuomotor policies and symbolic planning methods often fail to jointly handle continuous contact deviation and discrete industrial rule constraints. This paper proposes VTNS-Corr, a visual-tactile embodied perception-driven neuro-symbolic framework for rule consistency modeling and contact-deviation automatic correction. The method fuses stereo vision, tactile maps, force-torque feedback, proprioception, and human-safety cues into an embodied latent state, grounds it into differentiable symbolic predicates, evaluates fuzzy rule violations, attributes conflict sources, and generates corrective actions such as re-localization, grip adjustment, retreat-and-reinsert, and safety pause. Experiments on RLBench, ManiSkill2, RoboMimic, and DROID show that VTNS-Corr achieves the best assembly success rate, rule consistency score, correction success rate, and safety violation reduction. The results demonstrate its effectiveness for reliable and interpretable industrial humanoid precision assembly.
Industrial humanoid robots are increasingly expected to execute high-precision manipulation tasks, such as assembly, welding, and material handling, under dynamic and contact-rich working conditions. In such scenarios, operational safety and task reliability depend not only on accurate motion execution but also on timely detection of abnormal interaction patterns between the robot and its environment. This paper proposes VEDA (Visuo-Embodied Dual-Attention), a contact-aware multimodal framework for temporal anomaly detection that explicitly couples contact-guided spatial emphasis with stage-sensitive temporal modeling in industrial humanoid robots. VEDA integrates visual, force-aware, and motion-based embodied perception with a spatial-temporal dual-attention modeling mechanism. The spatial attention module emphasizes contact-relevant regions, while the temporal attention module captures critical operation stages and abnormal transitions over time. The core technical contribution lies in a dual-attention design in which force- and motion-related embodied cues guide where the model attends in space and when it attends in time during manipulation monitoring. By jointly modeling multimodal sensory observations and temporal dependencies, VEDA improves anomaly recognition, anomaly localization, and response timeliness during complex manipulation processes. Extensive experiments on benchmark manipulation datasets show that VEDA consistently outperforms representative baseline methods in anomaly detection accuracy, temporal precision, and robustness, demonstrating the effectiveness of task-oriented multimodal attention modeling for reliable industrial humanoid operation.
