The choice of driving mode will significantly affect the overall performance of the robot. Therefore, this article mainly focuses on the two mainstream driving technologies: hydraulic drive and motor drive. By analyzing the actual operational data of some typical products, we can understand their performance characteristics, core advantages, and inherent limitations, and then conduct multidimensional comparative analysis. This research result has the potential to provide a theory that can be directly used as a reference for system developers and mechanical designers.

Show more

Proximal Policy Optimization (PPO) is one of the most widely adopted reinforcement learning algorithms in both academic research and industrial applications. Its widespread popularity mainly comes from its outstanding training stability and simple implementation, which helps it avoid the common flaws of earlier policy gradient methods: unstable convergence and excessive sensitivity to hyperparameters. In recent years, the research community has put forward a large number of modified variants and targeted improvements to enhance PPO's learning efficiency, asymptotic performance and generalizability across different task environments. However, there is still a lack of systematic research on how the inherent complexity of a target task affects the actual effectiveness of these PPO improvements. In this paper, we try to fill this research gap by empirically explore how the task complexity would affect the effectiveness of PPO's improvements. This paper first conducted controlled evaluations of mainstream PPO variants in standard OpenAI Gym single-agent environments of explicitly graded difficulty, including low-complexity CartPole and medium-complexity LunarLander. The empirical results show that as task complexity increases, the performance reliability of PPO methods decrease significantly, and the performance gaps between different variants also widen noticeably. This paper further extend the analysis to multi-agent settings, which add extra challenges such as environmental non-stationarity and inter-agent coordination. Finally, this paper conclude that the effectiveness of PPO-based methods strongly depends on task complexity, which highlights that training stability and cross-environment adaptability are critical for developing PPO algorithms for complex and multi-agent scenarios.

Show more

There are three limitations in the research of anomaly detection for games, including reliance on single-dimensional features, inability to model heterogeneous relationships and failure to capture dynamic temporal evolution. To respond to these limitations, this paper proposes a model named DHGAD-Game, a dynamic heterogeneous graph anomaly detection for games. This model primarily consists of three modules. The heterogeneous subgraph construction module for relation-specific decomposition first decomposes the game dynamic graph into multiple heterogeneous subgraphs based on relation types. The temporal aware graph updating mechanism with GRU-based edge weight learning presents the core innovation of DHGAD-Game, which captures the dynamic evolution of entity interactions in gaming scenarios. The reconstruction-based anomaly detection module adopts autoencoder reconstruction error as the basis for anomaly scoring. Experimental results based on the CS2CD dataset demonstrate that the method proposed by this paper achieved a score of 0.561 and 0.816 in the metric of AUC-ROC and AUPRC, respectively. The results are significantly superior to some baseline methods such as AnticheatPT, Addgraph and TGN.

Show more

We're seeing more and more heterogeneous IoT devices being deployed acrossdifferent networks these days, and they're generating massive amounts of distributed data,this creates new challenges for secure multi-source aggregation in edge intelligenceenvironments. Most conventional approaches built for Digital Twin (DT) add way too muchmodeling complexity. They may not keep up when IoT environments shift quickly andunpredictably. Also, many existing homomorphic encryption solutions focus at privacyprotection, but they don't handle integrity verification during data aggregation processproperly. So to fix these issues, we developed a secure data aggregation framework for DigitalCousin (DC) systems. We built it using Paillier homomorphic encryption to keep data private,added RSA-based signatures so we can verify each data piece's authenticity, and a ChineseRemainder Theorem (CRT)-driven aggregation mechanism makes the actual aggregationprocess work smoothly. In this framework, edge gateways take charge of aggregatingencrypted data along with their associated signatures. Original plaintext stays completelyinaccessible throughout aggregation stage, no part of it can be accessed while the aggregationis still going on. Once the aggregated ciphertext is delivered to the DC side, signaturevalidation and plaintext reconstruction are carried out to guarantee both data authenticity andaggregation correctness. Compared with conventional aggregation schemes, the proposedframework decreases the processing pressure on edge gateways and reduces communicationcosts in large-scale IoT data collection tasks. The security evaluation further demonstratesthat the scheme can simultaneously support data confidentiality, integrity protection, andreliable source authentication in multi-device aggregation environments. Security analysisshows that the framework can effectively achieve confidentiality, integrity verification, andsource authentication for multi-device data aggregation scenarios.

Show more