Articles in this Volume

Research Article Open Access
Adaptive Motion Trajectory Control Method for Robots Based on Reinforcement Learning
Article thumbnail
Accurate manipulator trajectory control becomes difficult when nonlinear joint coupling, payload variation, and external disturbances invalidate fixed controller parameters. This study develops an adaptive trajectory controller based on Proximal Policy Optimization for a seven-degree-of-freedom Franka Emika Panda manipulator. The controller maps joint states, Cartesian tracking errors, target motion, and previous actions directly to continuous joint torques. Training is conducted in MuJoCo with randomized 0–2 kg payloads and 5–20 N external forces, while a composite reward simultaneously regulates Cartesian accuracy, orientation error, motion smoothness, control effort, and safety. Illustrative evaluation results show that PPO achieves a Cartesian RMSE of 6.84 ± 0.73 mm on the three-dimensional trajectory, compared with 13.27 ± 1.46 mm for computed-torque control and 21.63 ± 2.11 mm for PID. Under a 20 N disturbance, recovery time decreases to 0.41 ± 0.06 s, while trajectory jerk remains 31.7 ± 3.8 m/s³. The results indicate that dynamic randomization and multi-objective reward shaping enable the learned controller to preserve tracking accuracy and smoothness under changing robot dynamics.
Show more
Read Article PDF
Cite
Research Article Open Access
Literature Review: How Wearable Device Intervene Overweight/Obese People Loss Weight?
Background: Overweight and obesity present a significant global health challenge, associated with increased chronic disease risk and healthcare burden. Wearable devices, capable of continuously monitoring health metrics like activity and heart rate, offer a promising strategy for weight management interventions in this population. Objective: This review aimed to systematically elucidate the specific technological pathways, particularly the key sensors and devices employed, through which wearable devices intervene to facilitate weight loss in overweight or obese individuals. Method: A systematic search was conducted across major databases (PubMed, Scopus, Embase, Web of Science) for literature published between July 2015 and July 2025. Included studies detailed wearable device technical specifications (sensors, principles, algorithms), involved overweight/obese participants (BMI ≥25 kg/m²), and reported objective weight loss outcomes linked to device functionality. Screening yielded 14 eligible studies for analysis. Result: Wearable interventions primarily utilized commercial health trackers (e.g., Fitbit) or dedicated wearable sensors to monitor parameters like steps and heart rate, often combined with behavioral support. These devices facilitated weight loss through continuous monitoring, feedback provision, and behavioral prompting, demonstrating efficacy in reducing BMI, body weight, and visceral fat. However, adherence challenges, influenced by factors like age and technical issues, were identified as significant barriers to effective device utilization. Conclusion: Wearable devices represent a feasible intervention for weight loss in obese populations, primarily functioning via monitoring and feedback using commercial trackers or sensors. While effective, establishing clear criteria for device adherence and further investigating their specific contribution to weight loss outcomes are crucial future needs. Current evidence warrants cautious interpretation due to methodological limitations in existing studies.
Show more
Read Article PDF
Cite
Research Article Open Access
Warehouse Mobile Manipulation with Large Language Models and Foundation Models
Article thumbnail
Mobile manipulators can move between storage locations and handle goods, but transferring the semantic capabilities of foundation models to dependable warehouse operations remains constrained by execution interfaces, data coverage, and operating conditions. This thematic critical review examines perception, reasoning, planning and execution, and data and generalization. It compares hierarchical approaches with vision-language-action models and assesses their relevance to warehouse tasks using representative studies. Its distinctive value lies in relating model interfaces and transfer limits to warehouse order fulfillment, where stock-keeping unit (SKU) verification, shelf access, and onboard resource limits must be considered together. The evidence indicates that foundation models reorganize selected interfaces between observations, task descriptions, and actions; they do not remove the need for geometric feasibility checks, low-level control, or failure recovery. Open-vocabulary recognition must be connected to item-level verification, while semantic generalization should be distinguished from the acquisition of new motor skills. Reported transfer gains also depend on the task, embodiment, and adaptation budget. The review identifies order completion, coordinated base-arm motion, and measurable adaptation costs as priorities for research and evaluation. These conclusions support method selection and experimental design rather than a claim of validated warehouse deployment.
Show more
Read Article PDF
Cite
Research Article Open Access
Applications, Risks, and Future Prospects of Artificial Intelligence in Game Design--From Commercial Game Development to Serious Games
Nowadays, artificial intelligence (AI) no longer merely plays traditional roles in behavioral control, procedural generation, and data analysis in the game industry. It has become a production tool capable of generating images, code, and other interactive content. It allows game studios to shorten the time cost in development and lower the cost of prototype production. It also allows their games to deliver personalized player experiences. At the same time, AI also raises many concerns about copyright ownership and job displacement. Based on GDC surveys from the past three years, this paper first explores the current applications of AI in game industry and the controversies about its use. It then discusses several new directions of future applications, such as intelligent NPCs, dynamic storytelling, and personalized experiences. Finally, this paper explores the social value and ethical boundaries of AI in serious games related to social issues. This paper argues that the most appropriate role of AI in the game industry is not to replace game designers, but to play as a tool governed by design goals and clear accountability mechanisms. Only by maintaining a balance among efficiency, creativity, and transparency can AI truly improve the quality of game design.
Show more
Read Article PDF
Cite
Research Article Open Access
A Review of Five Major Machine Learning Paradigms and Their Representative Algorithms
Machine learning methods use different forms of data and feedback, which can make their relationships difficult for new learners to understand. This paper draws on related research to compare supervised, unsupervised, semi-supervised, self-supervised, and reinforcement learning and their representative algorithms. Foundational studies are considered alongside selected research from 2022–2024. Self-supervised learning is treated separately for clarity, while its relationship to unsupervised learning is acknowledged. The comparison considers learning signals, objectives, applications, strengths, and limitations. The main finding is that the five paradigms differ in how training information guides learning, rather than only in whether labels exist. Supervised learning supports prediction with known targets, while unsupervised learning supports grouping and exploration without such targets. Semi-supervised learning uses limited labels together with unlabeled examples, but its results depend on the training examples and the reliability of generated labels. Self-supervised learning creates targets from data and supports later tasks, although reduced labeling does not necessarily mean reduced computation. Reinforcement learning addresses decisions whose effects continue over time but learning stability and interaction costs require attention. The reviewed methods also show that a system can combine several paradigms. These findings suggest that method selection should begin with the task, available feedback, and evaluation conditions. No common ranking across all five paradigms follows from the reviewed evidence.
Show more
Read Article PDF
Cite
Research Article Open Access
The Comparison of Multi-Armed-Bandit Algorithms in Movie Recommendation Systems
Article thumbnail
With the development of information technology, especially the Internet, people are rapidly moving from the era of information scarcity to the era of information overload. Recommendation systems have thus rapidly developed, which aims to find the optimization of the experience for users. The Multi-Armed-Bandit (MAB) algorithms belong to the category of reinforcement learning , always trying to optimize the balance between exploration and exploitation trade-offs. This fundamental principle successfully coincides the application of recommendation systems. By using a large number of real user ratings of movies as the data-set, this paper compares the performances of various MAB algorithms, ε-greedy, Explore-Then-Commit (ETC), Upper Confidence Bound (UCB), Asymptotically optimal UCB and Thompson Sampling (TS) in the movie recommendation system. Through the performance of different algorithms on the same data, it is concluded that the TS algorithm performs relatively optimally in improving the efficiency of the system. Since the basic principle of TS algorithm utilizes Bayesian inference, the balance of each selection trade-off can more probably be better than other algorithms. TS algorithm has a strong flexibility and can perform a better balance when facing different environments. In addition, the methodological framework of this paper can be drawn upon in more complex scenarios and environments.
Show more
Read Article PDF
Cite