Learning paths personalized by behavior have become common among personalized learning systems. However, the traditional methods are usually inefficient at building coherent paths as the learner's knowledge state changes dynamically, and the prerequisite structure exists. In light of this problem, we present a generative learning-path optimization framework based on EdNet-KT1 interaction dataset, Transformer knowledge tracing model, Qwen2.5-7B-Instruct candidate generator, and constraint-aware reranking. Mastery profiles of learners are extracted, and then five ten-resource candidate paths are generated and sorted according to learning gain, prerequisites, difficulty deviation, and repetition. We compare our method to collaborative filtering, GRU4Rec, and knowledge-tracing greedy recommendation on offline experiments. The generative optimizer gets NDCG@10 of 0.589, future-concept coverage of 0.721, and prerequisite violation rate of 0.031, and thus outperforms the best baseline on all three metrics. Ablation analysis shows that the prerequisite violation rate rises to 0.091 without constraints, and the learning gain drops to 0.104 without the learning gain term, respectively. Moreover, the full model still achieves better NDCG@10 performance on all learner history categories than other baselines, costing 0.84 s per learner with the memory peak of 13.6 GB. This research demonstrates the effectiveness of combining the generation of learning paths with pedagogical constraints.

Autonomous all-terrain vehicles (ATVs) must be able to generate paths and maintain stable, executable vehicle motion in the face of varying, unknown conditions in off-road environments. This paper provides an overview of recent advances in terrain-aware motion modeling and autonomous path planning for ATVs. The first part of the manuscript discusses how characteristics of terrain such as slope, roughness, loose surfaces, vegetation, and uncertainty will impact slip, steering, vibrations, stability, and control of a vehicle. More recent studies rely on learning-based approaches, and uncertainty and motion are integrated into models of the terrain–vehicle interaction. The second part of manuscript focuses on planning methods for unknown environments.It discusses graph search,sampling-based approaches,and hierarchical planner,while also showing how these methods connect route selection with vehicle motion.In many cases,factors such as stability,uncertainly,and traversability are considered directly during the planning process. The planning methods reviewed in the manuscript integrate perception and control, actively leveraging the interaction with the environment. Planning methods have moved toward closed-loop control and interaction-based navigation. The integration of planning methods across vehicles, terrain types, and sensing and operational conditions will be the focus of future research to reduce planning complexity while ensuring planning and control methods operate safely and efficiently. This integration is important because it allows autonomous vehicles to operate more safety and efficiently in off-road environments that are both unfamiliar and constantly changing.

The rapid iterative development of generative AI technology is highly dependent on the supply of massive training data. The industry generally uses crawlers to grab Internet copyright works on a large scale for pre-training, which has led to global copyright disputes. The risk of copyright infringement in the acquisition and use of training data has become the core challenge restricting the development of industrial compliance. This article focuses on the complete life cycle of training data and explores three cutting-edge technological paths: pre-filtering of training data, in-process training constraints, and post-detection and evidence collection. Analyze the underlying operating principles of various technologies, conducts horizontal quantitative comparisons of mainstream technologies from four dimensions: detection accuracy, computing power cost, black-box adaptability, and judicial evidence collection effectiveness, and clarifies the practical bottlenecks of the existing protection system. The multi-layer technology defense and intellectual property system collaborative integrated protection framework is constructed in this article. It can achieve full process risk management of training data, reduce the probability of copyright infringement of large models, improve the efficiency of infringement tracing and evidence collection, and balance the protection of creator rights and the needs of AI technology innovation and development. This study can provide a complete theoretical basis for the construction of data compliance pipelines for large model enterprises, the development of industry compliance technical standards, and the improvement of digital supervision of intellectual property rights.

Exoskeleton robotics technology is a new kind of wearable bionic device that has been applied in various fields, such as healthcare and industry. However, there are still many development bottlenecks in the technology due to market and core-technology problems. The development history of exoskeleton robot technology, its applications in medical rehabilitation, outdoor walking assistance and industrial production, problems and future development directions of this technology are introduced in this paper. This paper studies the concept of passive exoskeletons and the development of the Hardiman project in the 1960s, and also analyzes the current progress in core technology and the emergence of commercial products. The first is to introduce the technical features of some typical products and compare differences in driving modes, intention recognition, safety and application subjects for exoskeleton robots in different use cases. Finally, this paper will summarize the current problems in technology for cost control, energy supply and human-computer interaction, and look ahead to the future development direction of new materials, integration of artificial intelligence and multi-area expansion

Humanoid robots require both high-performance drive structures and effective whole-body coordinated control to achieve dynamic motion, load transportation, and complex whole-body movements. This paper reviews the relationship between mechanical drive design and coordinated control methods for humanoid robots, with particular emphasis on the constraints imposed by actuator torque, mass distribution, transmission characteristics, and real-robot dynamics. At the hardware level, high-torque-density modular joints, lightweight transmissions, hybrid serial-parallel mechanisms, and drive-layout optimisation are examined in terms of their effects on joint torque, moving mass, and motion capability. At the control level, multi-joint coordinated motion control, model predictive control, reinforcement learning, and dynamics adaptation methods are discussed for maintaining stability and improving actuator utilisation under contact and joint constraints. The review further considers the differences between idealised simulation models and physical robots caused by friction, hysteresis, drive-torque deviations, contact compliance, and communication delays. The analysis indicates that drive structures establish the hardware boundaries of humanoid motion capability, while whole-body coordinated control determines how effectively these capabilities can be realised during operation. The paper therefore highlights the importance of integrated electromechanical and control co-design, including joint parameters, drive layouts, mass distribution, state estimation, and dynamics adaptation, as a potential direction for improving the robustness and efficiency of humanoid robots.